BIM-assisted steel structure construction safety management early warning method, device and equipment

Abstract

The application relates to the technical field of data processing, and discloses a BIM-assisted steel structure construction safety management early warning method, device and equipment. The method comprises the following steps: collecting multi-source heterogeneous data of a steel structure construction site and performing space-time alignment processing to obtain engineering data flow; performing dynamic noise reduction and feature extraction on the engineering data flow to obtain a standardized multi-modal feature vector and performing time sequence correlation analysis and pattern mining to obtain a potential risk index set; performing multi-level threshold calculation and context-sensitive evaluation on the potential risk index set to obtain an adaptive early warning trigger matrix; performing multi-dimensional cross-validation and causal chain analysis on the adaptive early warning trigger matrix to obtain a graded and classified risk assessment report; and performing scene analysis and response scheme generation on the graded and classified risk assessment report to obtain an executable multi-step emergency plan. The application improves the efficiency and accuracy of BIM-assisted steel structure construction safety management early warning.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a BIM-assisted method, device and equipment for early warning of safety management in steel structure construction. Background Technology

[0002] Steel structures are widely used in modern buildings and infrastructure due to their advantages such as high strength, lightweight, and rapid construction. Traditional steel structure construction safety management mainly relies on manual inspections, periodic testing, and experience-based judgment. With the development of information technology, advanced monitoring methods such as strain sensors and inclinometers have been introduced into steel structure construction safety monitoring. Simultaneously, the application of Building Information Modeling (BIM) technology has provided new possibilities for information management throughout the entire lifecycle of steel structures. Some studies are attempting to combine BIM with Internet of Things (IoT) technology to achieve real-time monitoring and visualized management of the steel structure construction process.

[0003] However, existing methods for safety management in steel structure construction still have some shortcomings. First, most methods focus only on single or a few risk factors, making it difficult to comprehensively consider the combined impact of multi-source heterogeneous data. Second, existing early warning systems often use fixed thresholds, lacking adaptability to complex construction environments. Third, risk assessments mostly remain at the qualitative analysis level, failing to provide accurate risk quantification results. Finally, current emergency plans are usually pre-defined static schemes, unable to be dynamically adjusted according to real-time conditions. These problems limit the effectiveness and accuracy of safety management in steel structure construction. Summary of the Invention

[0004] This application provides a BIM-assisted method, device, and equipment for early warning of safety management in steel structure construction, which improves the efficiency and accuracy of BIM-assisted early warning of safety management in steel structure construction.

[0005] Firstly, this application provides a BIM-assisted early warning method for safety management in steel structure construction. The method includes: collecting multi-source heterogeneous data from the steel structure construction site and performing spatiotemporal alignment processing to obtain a high-dimensional synchronized engineering data stream; performing dynamic noise reduction and feature extraction on the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector; performing temporal correlation analysis and pattern mining on the standardized multimodal feature vector to obtain a set of potential risk indicators; performing multi-level threshold calculation and context-sensitive assessment on the set of potential risk indicators to obtain an adaptive early warning trigger matrix; performing multi-dimensional cross-validation and causal chain analysis on the adaptive early warning trigger matrix to obtain a graded and categorized risk assessment report; and performing scenario-based analysis and response scheme generation on the graded and categorized risk assessment report to obtain an executable multi-step emergency plan.

[0006] Secondly, this application provides a BIM-assisted steel structure construction safety management early warning device, the BIM-assisted steel structure construction safety management early warning device comprising:

[0007] The processing module is used to collect and spatiotemporally align multi-source heterogeneous data from the steel structure construction site to obtain a high-dimensional synchronized engineering data stream.

[0008] The extraction module is used to perform dynamic noise reduction and feature extraction on the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector;

[0009] The mining module is used to perform time-series correlation analysis and pattern mining on the standardized multimodal feature vectors to obtain a set of potential risk indicators;

[0010] The evaluation module is used to perform multi-level threshold calculation and context-sensitive evaluation on the potential risk indicator set to obtain an adaptive early warning trigger matrix;

[0011] The analysis module is used to perform multi-dimensional cross-validation and causal chain analysis on the adaptive early warning trigger matrix to obtain a graded and classified risk assessment report;

[0012] The generation module is used to perform scenario-based analysis and response plan generation on the risk assessment reports of the hierarchical classification, so as to obtain an executable multi-step emergency plan.

[0013] A third aspect of this application provides a computer device in which the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via a bus, and when the machine-readable instructions are executed by the processor, the steps of the BIM-assisted steel structure construction safety management early warning method described above are performed.

[0014] The technical solution provided in this application obtains a high-dimensional synchronized engineering data stream by acquiring multi-source heterogeneous data from the steel structure construction site and performing spatiotemporal alignment processing. This enables comprehensive and multi-angle real-time monitoring of the construction site, providing a complete and accurate data foundation for subsequent analysis. Secondly, dynamic noise reduction and feature extraction are performed on the high-dimensional synchronized engineering data stream to obtain standardized multimodal feature vectors. This effectively eliminates noise interference in the data, extracts the most representative features, and improves the accuracy and efficiency of subsequent analysis. Thirdly, by performing temporal correlation analysis and pattern mining on the standardized multimodal feature vectors, a potential risk indicator set is obtained. This deeply explores the implicit time dependencies and risk patterns in the data, providing a scientific basis for risk early warning. Furthermore, multi-level threshold calculations and context-sensitive assessments are performed on the potential risk indicator set to obtain an adaptive early warning trigger matrix. This enables dynamic adjustment of the early warning mechanism, significantly improving the accuracy and adaptability of early warnings. Simultaneously, by performing multi-dimensional cross-validation and causal chain analysis on the adaptive early warning trigger matrix, a graded and classified risk assessment report is obtained. This not only quantifies the degree of risk but also reveals the causal relationships between risk factors, providing comprehensive support for risk management decisions. Finally, the risk assessment reports, categorized by level and type, are analyzed in a scenario-based manner to generate response plans, resulting in an executable multi-step emergency plan. This achieves intelligent management of the entire process from risk identification to emergency response. This method greatly improves the accuracy, real-time performance, and operability of steel structure construction safety management, effectively reduces the probability of safety accidents, and enhances the efficiency and effectiveness of emergency response. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of an embodiment of the BIM-assisted steel structure construction safety management early warning method in this application.

[0017] Figure 2 This is a schematic diagram of one embodiment of the BIM-assisted steel structure construction safety management and early warning device in this application.

[0018] Figure 3 This is a schematic diagram of the structure of the computer device in the embodiments of this application. Detailed Implementation

[0019] This application provides a BIM-assisted method, apparatus, and equipment for safety management and early warning in steel structure construction. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0020] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the BIM-assisted steel structure construction safety management early warning method in this application includes:

[0021] Step S101: Perform multi-source heterogeneous data acquisition and spatiotemporal alignment processing on the steel structure construction site to obtain a high-dimensional synchronized engineering data stream;

[0022] Step S102: Perform dynamic noise reduction and feature extraction on the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector;

[0023] Step S103: Perform time-series correlation analysis and pattern mining on the standardized multimodal feature vectors to obtain a set of potential risk indicators;

[0024] Step S104: Perform multi-level threshold calculation and context-sensitive assessment on the potential risk indicator set to obtain the adaptive early warning trigger matrix;

[0025] Step S105: Perform multi-dimensional cross-validation and causal chain analysis on the adaptive early warning trigger matrix to obtain a graded and classified risk assessment report;

[0026] Step S106: Analyze the risk assessment report of the classification and categorization in a scenario-based manner and generate response plans to obtain an executable multi-step emergency plan.

[0027] It is understood that the executing entity of this application can be a BIM-assisted steel structure construction safety management and early warning device, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0028] Specifically, multi-source heterogeneous data acquisition and spatiotemporal alignment processing are performed on the steel structure construction site to obtain a high-dimensional synchronized engineering data stream. This involves the integration of multiple data sources, including point cloud data acquired by 3D laser scanning, real-time strain data acquired by strain sensors, multi-dimensional environmental parameter data recorded by environmental monitoring equipment, and dynamic position information captured by a real-time positioning system. By performing spatiotemporal alignment processing on these data, the consistency of data from different sources in time and space is ensured, thereby forming a high-dimensional synchronized engineering data stream. Dynamic denoising and feature extraction are then performed on the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector. This step first applies adaptive wavelet transform to decompose the data at multiple scales, and then eliminates the influence of noise through threshold denoising. Subsequently, dimensionality reduction is performed using principal component analysis, and high-order feature representations are obtained through nonlinear mapping. Key frequency features are extracted using methods such as time-frequency domain transformation and peak detection, and finally, standardized multimodal feature vectors are obtained through statistical moment extraction and normalization.

[0029] A set of potential risk indicators is obtained by performing time-series correlation analysis and pattern mining on standardized multimodal feature vectors. This step first decomposes the feature vectors into time series components, separating trend, periodic, and random components, and obtains time-dependent characteristics through autocorrelation analysis. Dynamic time warping is used to align the feature sequences, followed by sliding window segmentation to obtain local feature fragments. Frequent pattern mining and association rule learning are used to identify recurring feature combinations and association rules between features. Finally, graph structure modeling and community detection are used to identify highly correlated feature clusters and score their importance, resulting in a set of potential risk indicators. Multi-level threshold calculation and context-sensitive assessment are then performed on the potential risk indicator set to obtain an adaptive early warning trigger matrix. This step first performs historical data statistical analysis on the indicator set to calculate multi-level quantiles to obtain an initial threshold set. The threshold range is adjusted by incorporating expert knowledge, and multi-level fuzzy thresholds are obtained through fuzzy set partitioning. Dynamic time windows and contextual information are applied to obtain context-sensitive thresholds. A threshold decision tree is formed through multi-dimensional cross-combination, and simplified decision rules are obtained through pruning optimization. Finally, the decision rules are mapped to an early warning trigger condition matrix, and an adaptive adjustment mechanism is designed to obtain an adaptive early warning trigger matrix.

[0030] The adaptive early warning trigger matrix undergoes multi-dimensional cross-validation and causal chain analysis to obtain a graded and categorized risk assessment report. This step first identifies similar early warning scenarios through historical case matching and performs feedback analysis to evaluate the accuracy of the early warnings. Multi-fold cross-validation assesses the stability of the early warnings, and sensitivity analysis identifies key influencing factors. A causal inference model is constructed to obtain a potential causal relationship network, and path analysis determines the risk propagation chain. Multiple possible risk evolution paths are generated through scenario simulation, and probability assessment is performed to obtain risk level classifications. Finally, a multi-dimensional comprehensive score is applied to the risk levels to form a risk severity matrix, which is then structurally described to obtain a graded and categorized risk assessment report. Finally, the graded and categorized risk assessment report undergoes scenario-based analysis and response plan generation to obtain an executable multi-step emergency plan. This step first performs semantic parsing of the risk assessment report, extracts key risk descriptors, and constructs a risk scenario model. Available emergency resources are analyzed, and resource allocation strategies are formulated. Response measures are decomposed into a sequence of preliminary action steps, and temporal logic is optimized. Responsibility roles are assigned to each action step, and an information transmission network is designed. Finally, key decision nodes are identified, alternative solutions are generated, and an executable multi-step emergency plan is formed.

[0031] For example, in a steel structure construction project, a high-dimensional engineering data stream containing 100,000 data points was obtained through multi-source data acquisition. After dynamic noise reduction and feature extraction, the data dimension was reduced from the original 100 dimensions to 20 dimensions, forming a standardized multimodal feature vector. Temporal correlation analysis and pattern mining identified 15 potential risk indicators, including key parameters such as structural stress, ambient temperature, and wind speed. Multi-level threshold calculation set three warning thresholds for each indicator, forming 45 warning trigger conditions. Cross-validation showed that the warning accuracy rate reached 85%, and causal chain analysis plotted three main risk propagation paths. The final emergency plan included five main response phases, involving 12 specific execution steps, and clearly defined the responsibilities of seven key roles.

[0032] In this embodiment, multi-source heterogeneous data acquisition and spatiotemporal alignment processing of the steel structure construction site yields a high-dimensional synchronized engineering data stream, enabling comprehensive and multi-angle real-time monitoring of the construction site and providing a complete and accurate data foundation for subsequent analysis. Secondly, dynamic noise reduction and feature extraction are performed on the high-dimensional synchronized engineering data stream to obtain standardized multimodal feature vectors, effectively eliminating noise interference in the data and extracting the most representative features, thus improving the accuracy and efficiency of subsequent analysis. Thirdly, temporal correlation analysis and pattern mining are conducted on the standardized multimodal feature vectors to obtain a potential risk indicator set, deeply exploring the implicit time dependencies and risk patterns in the data, providing a scientific basis for risk warning. Furthermore, multi-level threshold calculation and context-sensitive assessment are performed on the potential risk indicator set to obtain an adaptive warning trigger matrix, realizing dynamic adjustment of the warning mechanism and significantly improving the accuracy and adaptability of the warning. Simultaneously, multi-dimensional cross-validation and causal chain analysis are performed on the adaptive warning trigger matrix to obtain a graded and classified risk assessment report, which not only quantifies the degree of risk but also reveals the causal relationships between risk factors, providing comprehensive support for risk management decisions. Finally, the risk assessment reports, categorized by level and type, are analyzed in a scenario-based manner to generate response plans, resulting in an executable multi-step emergency plan. This achieves intelligent management of the entire process from risk identification to emergency response. This method greatly improves the accuracy, real-time performance, and operability of steel structure construction safety management, effectively reduces the probability of safety accidents, and enhances the efficiency and effectiveness of emergency response.

[0033] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0034] (1) Perform three-dimensional laser scanning on the steel structure components to obtain high-precision point cloud data, and perform noise filtering and sparsification on the high-precision point cloud data to obtain structural geometric contour data.

[0035] (2) Feature extraction and parametric modeling of the structural geometric contour data are performed to obtain the steel structure BIM model, and attribute information is added to the steel structure BIM model to obtain the semantic BIM model.

[0036] (3) Strain sensors are arranged at key nodes of the steel structure to obtain real-time strain data, and signal denoising and outlier detection are performed on the real-time strain data to obtain an effective strain dataset.

[0037] (4) Deploy environmental monitoring equipment at the construction site to obtain multidimensional environmental parameter data, and perform data standardization and interpolation on the multidimensional environmental parameter data to obtain a continuous environmental state sequence;

[0038] (5) Configure the real-time positioning system for construction personnel and equipment to obtain dynamic location information, and perform trajectory smoothing and predictive analysis on the dynamic location information to obtain spatiotemporal behavior pattern data.

[0039] (6) The semantic BIM model, effective strain dataset, continuous environmental state sequence and spatiotemporal behavior pattern data are timestamped to obtain a multi-source data synchronization matrix. The multi-source data synchronization matrix is ​​then spatially registered and scaled to obtain a high-dimensional synchronized engineering data stream.

[0040] Specifically, 3D laser scanning is performed on steel structure components to acquire high-precision point cloud data. 3D laser scanning technology rapidly captures the 3D coordinate information of an object's surface by emitting laser light and receiving reflected signals, forming dense point cloud data. These raw point cloud data undergo noise filtering and sparsification to remove outliers and redundant information, resulting in clearer structural geometric contour data. Noise filtering typically employs statistical outlier removal algorithms, while sparsification uses methods such as voxel meshing or octree segmentation to reduce data volume while preserving key geometric features. Feature extraction and parametric modeling are then performed on the structural geometric contour data to construct a steel structure BIM model. Feature extraction includes identifying basic geometric elements such as planes, edges, and corners, as well as structural components such as beams, columns, and nodes. Parametric modeling, based on these extracted features, uses a predefined parametric component library to quickly generate an editable 3D model. Furthermore, attribute information is supplemented to the steel structure BIM model, including material properties, construction progress, safety regulations, and other multi-dimensional information, resulting in a semantic BIM model. This semantic model not only contains geometric information but also rich semantic information, providing comprehensive data support for subsequent security management and risk analysis.

[0041] Strain sensors are deployed at key nodes of the steel structure to collect strain data in real time. Strain sensors reflect the structural stress state by measuring material deformation and are a crucial means of monitoring structural safety. The collected real-time strain data undergoes signal denoising and outlier detection to improve data quality. Signal denoising commonly uses methods such as wavelet transform or Kalman filtering, while outlier detection can employ statistical or machine learning-based methods, such as the 3σ criterion or the isolated forest algorithm. The resulting effective strain dataset is more reliable and accurately reflects the actual stress state of the structure. Simultaneously, environmental monitoring equipment is deployed at the construction site to acquire multi-dimensional environmental parameter data. This data includes key environmental factors affecting construction safety, such as temperature, humidity, wind speed, noise, and dust concentration. These raw data are standardized to eliminate dimensional differences between parameters, facilitating comprehensive analysis. Furthermore, interpolation, such as linear interpolation or spline interpolation, fills in missing values ​​that may have appeared during data acquisition, obtaining a continuous environmental state sequence, providing a complete data foundation for subsequent time-series analysis and risk prediction.

[0042] To monitor the dynamic situation at the construction site, a real-time positioning system is configured for construction personnel and equipment to acquire dynamic location information. This can be achieved through technologies such as RFID, GPS, or UWB. The acquired raw location data undergoes trajectory smoothing to eliminate positioning errors and jumps, resulting in more coherent movement trajectories. Predictive analysis is then performed, using methods such as Kalman filtering or Long Short-Term Memory (LSTM) networks to predict future location changes. These processes yield spatiotemporal behavioral pattern data reflecting the movement patterns of personnel and equipment, providing crucial reference for safety management. Finally, the semantic BIM model, effective strain dataset, continuous environmental state sequence, and spatiotemporal behavioral pattern data are timestamped to ensure consistency across different data sources in the time dimension, forming a multi-source data synchronization matrix. This matrix is ​​then spatially registered and scaled to address potential coordinate system differences and scale inconsistencies between different data sources. Spatial registration typically employs feature point matching or Iterative Closest Point (ICP) algorithms, while scale unification is achieved through coordinate transformation. These processes ultimately yield a high-dimensional synchronized engineering data stream, providing a comprehensive and consistent data foundation for subsequent safety risk analysis.

[0043] For example, in a steel structure engineering project, 3D laser scanning acquired raw point cloud data containing 100 million points. After noise filtering and sparsification, the data volume was reduced to 20 million points while retaining key structural geometric information. Based on this data, a steel structure BIM model containing 500 components was constructed, and 2,000 attribute information were added to form a semantic BIM model. Fifty strain sensors were deployed at key nodes, collecting data 100 times per second. After noise reduction and outlier detection, 95% of the valid data was retained. Environmental monitoring equipment recorded 10 environmental parameters per minute, and through standardization and interpolation, a continuous 24-hour environmental state sequence was generated. A real-time positioning system tracked the location information of 100 workers and 20 pieces of equipment, and after trajectory smoothing and predictive analysis, 8 hours of spatiotemporal behavior pattern data were obtained. Finally, these multi-source heterogeneous data were integrated into a high-dimensional synchronized engineering data stream containing 10,000 time points and 1,000 feature dimensions, providing a comprehensive and unified data foundation for subsequent safety risk analysis. Through the collection, processing, and fusion of this multi-source data, comprehensive and real-time monitoring of the steel structure construction site was achieved.

[0044] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0045] (1) Perform adaptive wavelet transform on the high-dimensional synchronized engineering data stream to obtain multi-scale decomposition results, and perform threshold denoising on the multi-scale decomposition results to obtain smoothed data stream.

[0046] (2) Perform principal component analysis on the smoothed data stream to obtain a dimensionality-reduced feature set, and perform nonlinear mapping on the dimensionality-reduced feature set to obtain a higher-order feature representation;

[0047] (3) Perform time-frequency domain transformation on the higher-order feature representation to obtain the time-frequency feature matrix, and perform peak detection and spectral analysis on the time-frequency feature matrix to obtain the key frequency features;

[0048] (4) Statistical moments are extracted from key frequency features to obtain statistical feature vectors, and the statistical feature vectors are normalized to obtain standardized feature sets.

[0049] (5) Perform multimodal fusion on the standardized feature set to obtain the fused feature tensor, and perform tensor decomposition on the fused feature tensor to obtain the standardized multimodal feature vector.

[0050] Specifically, adaptive wavelet transform can automatically adjust wavelet basis functions according to the local characteristics of the data to achieve multi-scale decomposition of signals. This process decomposes the original data into sub-signals at different frequency levels, effectively separating noise and useful information. Threshold denoising is then applied to the obtained multi-scale decomposition results. By setting an appropriate threshold, significant coefficients are retained while small-amplitude coefficients are suppressed, resulting in a smoothed data stream. This method effectively removes high-frequency noise while preserving the main features of the signal. Principal component analysis (PCA) is then performed on the smoothed data stream to achieve dimensionality reduction. PCA projects high-dimensional data into a low-dimensional space by identifying the main directions of change in the data, obtaining a dimensionality-reduced feature set. This step not only reduces data redundancy but also extracts the most representative features from the data. To capture nonlinear relationships in the data, a nonlinear mapping is performed on the dimensionality-reduced feature set to obtain a higher-order feature representation. Commonly used methods for nonlinear mapping include kernel principal component analysis (KPCA) or deep learning techniques such as autoencoders, which can better express the inherent structure and complex patterns of the data.

[0051] Subsequently, a time-frequency domain transformation is performed on the higher-order feature representation to obtain the time-frequency feature matrix. Time-frequency domain transformations, such as Short-Time Fourier Transform (STFT) or wavelet transform, can simultaneously analyze the characteristics of the signal in the time and frequency domains, providing a more comprehensive signal characterization. Peak detection and spectral analysis are performed on the time-frequency feature matrix to identify significant frequency components and energy distributions in the signal, thereby obtaining key frequency features. These features reflect the response characteristics of the steel structure at different frequencies and are of great significance for identifying structural anomalies and potential risks. Statistical moment extraction is performed on the key frequency features to calculate statistical measures such as mean, variance, skewness, and kurtosis, obtaining statistical feature vectors. These statistical features provide important information about the data distribution, helping to capture the overall characteristics and anomaly patterns of the data. To eliminate dimensional differences between different features, the statistical feature vectors are normalized to obtain a standardized feature set. Normalization ensures that different features have the same weight in subsequent analysis, improving the effectiveness of feature fusion.

[0052] Finally, multimodal fusion is performed on the standardized feature set, integrating features from different data sources (such as BIM models, strain data, environmental data, etc.) to obtain a fused feature tensor. Multimodal fusion fully utilizes the complementarity of different types of data, providing a more comprehensive information representation. Tensor decomposition, such as Tucker decomposition or CP decomposition, is then performed on the fused feature tensor to further extract the core structure and correlation patterns of the multimodal data, ultimately yielding a standardized multimodal feature vector. This feature vector integrates information from multiple data sources, providing high-quality input for subsequent safety risk analysis.

[0053] For example, in a steel structure engineering project, a raw data stream containing 1000 time points and 100 sensors was collected. Adaptive wavelet transform was used to decompose the data into four scale levels, each containing sub-signals at 250 time points. Threshold denoising reduced noise amplitude by 80%, resulting in a smoothed data stream. Principal component analysis reduced the 100-dimensional features to 20 dimensions, retaining 95% of the variance information. Nonlinear mapping, using a three-layer autoencoder, mapped the 20-dimensional features to a 50-dimensional higher-order feature space. Time-frequency domain transformation used STFT with a window size of 64, resulting in a 64×250 time-frequency feature matrix. Peak detection identified 10 key frequency features. Statistical moment extraction calculated four statistics, yielding a 40-dimensional statistical feature vector. Normalization scaled all features to the range [-1,1]. Multimodal fusion integrated the 10-dimensional geometric features, 20-dimensional material features, and 5-dimensional features from the BIM model, forming a 75-dimensional fused feature tensor. Tensor decomposition ultimately yields a 50-dimensional standardized multimodal feature vector. This feature vector integrates information from multiple aspects, including structural geometry, material properties, dynamic response, and environmental factors, providing a comprehensive and accurate data foundation for the safety management of steel structure construction.

[0054] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0055] (1) Perform time series decomposition on the standardized multimodal feature vector to obtain trend, periodic and random components, and perform autocorrelation analysis on the trend, periodic and random components to obtain time dependence characteristics;

[0056] (2) Dynamic time warping is performed on the time-dependent characteristics to obtain the aligned feature sequence, and sliding window segmentation is performed on the aligned feature sequence to obtain local feature fragments;

[0057] (3) Frequent pattern mining is performed on local feature segments to obtain recurring feature combinations, and association rule learning is performed on the recurring feature combinations to obtain association rules between features;

[0058] (4) Graph structure modeling is performed on the association rules between features to obtain the feature association network, and community detection is performed on the feature association network to obtain highly correlated feature clusters;

[0059] (5) The importance of highly related feature clusters is scored to obtain key feature subsets, and the key feature subsets are semantically interpreted to obtain a set of potential risk indicators.

[0060] Specifically, time series decomposition of standardized multimodal feature vectors is a crucial step in in-depth analysis of data characteristics. Decomposing the original time series into trend, periodic, and random components reveals the inherent structure of the data. The trend component reflects the long-term trend of data changes, the periodic component captures the periodic fluctuations, and the random component represents irregular fluctuations and noise. Autocorrelation analysis is performed on these three components separately, calculating the autocorrelation coefficients at different time lags to obtain time dependence characteristics. Autocorrelation analysis reveals the correlation of data at different time scales, providing an important basis for subsequent risk prediction. Dynamic Time Warping (DTW) is then applied to the time dependence characteristics. The DTW algorithm, through nonlinear alignment of time series, can effectively handle time series with different rates or lengths, enabling similar patterns to be aligned. This step yields an aligned feature sequence, laying the foundation for subsequent pattern recognition. Then, a sliding window segmentation is applied to the aligned feature sequence to obtain local feature fragments. The sliding window technique divides a long time series into multiple short time fragments by moving a fixed-size window across the time series, facilitating the capture of local patterns and features.

[0061] Frequent pattern mining is performed on local feature fragments to identify recurring feature combinations. Frequent pattern mining algorithms, such as Apriori or FP-Growth, can discover frequently occurring feature sets from large amounts of data. These recurring feature combinations may represent typical states or potential risk patterns in the steel structure construction process. Subsequently, association rule learning is performed on these recurring feature combinations to obtain association rules between features. Association rule learning can not only discover strong correlations between features but also quantify the confidence and support of these correlations, providing quantitative evidence for risk assessment. Graph structure modeling is then performed on the association rules between features to construct a feature association network. In this network, nodes represent different features, and edges represent the correlation strength between features. This graph structure representation intuitively shows the complex relationships between features, facilitating subsequent analysis. Community detection is performed on the constructed feature association network to identify highly correlated feature clusters. Community detection algorithms, such as the Louvain method or spectral clustering, can group closely connected nodes together to form feature clusters. These feature clusters represent groups of interrelated risk factors in steel structure construction safety management.

[0062] Finally, the importance of highly correlated feature clusters is scored to obtain a subset of key features. Importance scoring can be based on various metrics, such as node degree centrality, betweenness centrality, or eigenvector centrality. These metrics measure the importance of features within the entire network from different perspectives. Based on the scoring results, the most important feature subset is selected, and these features are semantically interpreted. The data analysis results are then correlated with actual engineering concepts to ultimately obtain a set of potential risk indicators. These risk indicators directly reflect potential safety hazards during steel structure construction, providing clear direction for management decisions.

[0063] For example, in a large-scale steel structure project, standardized multimodal feature vectors containing 100 features and 10,000 time points were collected. Time series decomposition was used to break down each feature sequence into three components: trend, periodic, and random. Autocorrelation analysis of these components revealed that 50% of the features exhibited significant autocorrelation over a 24-hour period, reflecting the periodic impact of daily construction activities. Dynamic time warping aligned the different feature sequences, reducing the average alignment error from 2.5 hours to 0.5 hours. Using a 12-hour sliding window, approximately 2,000 local feature fragments were obtained. Frequent pattern mining identified 50 frequent feature combinations, including a typical combination such as "steel beam stress-temperature change-wind speed." Association rule learning generated 200 rules, such as "when the temperature rises by 5°C and the wind speed exceeds 10 m / s, the probability of a 20% increase in steel beam stress is 0.8." The feature association network contained 100 nodes and 500 edges, and the community detection algorithm identified 10 highly correlated feature clusters. Importance scoring, based on the PageRank algorithm, selected the top 20 most important features as a subset of key features. Semantic interpretation mapped these features to practical engineering concepts, such as "main beam deformation rate," "nodal stress concentration," and "environmental temperature fluctuations," ultimately forming 15 specific potential risk indicators. These indicators cover multiple aspects, including structural stress, environmental factors, and construction progress, providing a comprehensive and accurate risk assessment basis for steel structure construction safety management.

[0064] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0065] (1) Perform historical data statistical analysis on the potential risk indicator set to obtain the indicator distribution characteristics, and perform multi-level quantile calculation on the indicator distribution characteristics to obtain the initial threshold set;

[0066] (2) The initial threshold set is fused with expert knowledge to obtain the adjusted threshold range, and the adjusted threshold range is divided into fuzzy sets to obtain multi-level fuzzy thresholds.

[0067] (3) Apply dynamic time window to the multi-level fuzzy threshold to obtain a time-varying threshold sequence, and associate the time-varying threshold sequence with context information to obtain a context-sensitive threshold;

[0068] (4) Perform multi-dimensional cross-combination on context-sensitive thresholds to obtain a threshold decision tree, and perform pruning optimization on the threshold decision tree to obtain simplified decision rules;

[0069] (5) Perform matrix mapping on the simplified decision rules to obtain the early warning trigger condition matrix, and design an adaptive adjustment mechanism for the early warning trigger condition matrix to obtain the adaptive early warning trigger matrix.

[0070] Specifically, the probability distribution characteristics of each indicator are calculated, including statistics such as mean, standard deviation, skewness, and kurtosis. Then, multi-level quantile calculations are performed on these indicator distribution characteristics to obtain the initial threshold set. The multi-level quantile calculation can be expressed by the following formula:

[0071] Q p =F -1 (p)

[0072] Among them, Q p It is the p-quantile, F -1 It is the inverse function of the cumulative distribution function, where p is the probability value (e.g., 0.95, 0.99, etc.). This method allows setting multiple initial thresholds for each risk indicator with different warning levels. Next, expert knowledge is fused into the initial threshold set to obtain an adjusted threshold range. This step adjusts the statistically obtained thresholds using methods such as the Delphi method or the Analytic Hierarchy Process (AHP), combined with expert experience. The adjusted threshold range better aligns with practical engineering experience and safety management needs. Subsequently, the adjusted threshold range is partitioned into fuzzy sets to obtain multi-level fuzzy thresholds. Fuzzy set theory can better handle the uncertainty and transition of thresholds; fuzzy thresholds can be represented by membership functions.

[0073]

[0074] Where, μ A (x) represents the membership degree of element x to fuzzy set A, and a and b are the lower and upper thresholds, respectively. A dynamic time window is applied to the multi-level fuzzy thresholds to obtain a time-varying threshold sequence. The dynamic time window considers the influence of different time periods (such as seasons and day / night cycles) on the thresholds, allowing the thresholds to adjust dynamically over time. This can be achieved using a sliding window technique, with the window size determined based on the specific application scenario. Then, contextual information is correlated to the time-varying threshold sequence to obtain context-sensitive thresholds. Contextual information includes factors such as weather conditions, construction stage, and surrounding environment. These factors are incorporated through a weighted adjustment method to make the thresholds more closely reflect actual conditions.

[0075] A threshold decision tree is obtained by performing multi-dimensional cross-combinations on context-sensitive thresholds. Each node in the decision tree represents the threshold judgment of a risk indicator, and the path represents the combined judgment conditions of multiple indicators. To avoid overly complex decision trees, the threshold decision tree is pruned to obtain simplified decision rules. The pruning process can use indicators such as error rate or complexity penalty to improve decision efficiency while maintaining early warning accuracy. Finally, the simplified decision rules are matrix-mapped to obtain the early warning trigger condition matrix. This matrix clearly represents the early warning levels corresponding to different combinations of risk indicators. To enable the early warning system to adapt to changes in actual engineering conditions, an adaptive adjustment mechanism is designed for the early warning trigger condition matrix to obtain an adaptive early warning trigger matrix. The adaptive adjustment mechanism can dynamically update the trigger conditions in the matrix based on early warning effect feedback, thereby continuously optimizing early warning performance.

[0076] For example, in a large-scale steel structure project, 10 key risk indicators were selected, including main beam deflection, nodal stress, and ambient temperature. Statistical analysis of three years of historical data was used to calculate the distribution characteristics of each indicator. Taking main beam deflection as an example, its mean was 20 mm and its standard deviation was 2 mm. Using multi-level quantiles, three warning thresholds were obtained: mild warning (95th quantile, 24 mm), moderate warning (99th quantile, 26 mm), and severe warning (99.9th quantile, 28 mm). During the expert knowledge fusion phase, five senior engineers adjusted the initial thresholds using the Delphi method. For example, considering safety margins, the severe warning threshold for main beam deflection was adjusted to 27 mm. Fuzzy set partitioning transformed this threshold into a fuzzy interval, such as [26 mm, 28 mm], within which the warning level gradually increases with increasing deflection.

[0077] The dynamic time window application considers the impact of diurnal temperature variations on structural deformation, slightly relaxing the threshold at night (e.g., increasing it by 1 mm). Contextual information association further considers the impact of wind loads, lowering the threshold by 5% during windy weather (wind speeds exceeding 10 m / s). The decision tree formed by multi-dimensional cross-combinations contains over 100 nodes, which are reduced to 50 key nodes through pruning optimization, resulting in simplified decision rules. The final early warning trigger condition matrix is ​​a 10×3 matrix (10 indicators, 3 early warning levels), where the matrix elements are the specific thresholds for each indicator.

[0078] The adaptive adjustment mechanism employs a feedback loop to evaluate the early warning effectiveness weekly. If the false alarm rate is found to be too high (e.g., exceeding 10%), the thresholds for relevant indicators are appropriately relaxed; conversely, if the false alarm rate is too high (e.g., exceeding 1%), the thresholds are tightened accordingly. In this way, the early warning system can continuously self-optimize, adapt to changes in actual engineering conditions, and improve the accuracy and reliability of early warnings.

[0079] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0080] (1) Perform historical case matching on the adaptive early warning trigger matrix to obtain a set of similar early warning scenarios, and perform result feedback analysis on the set of similar early warning scenarios to obtain an assessment of early warning accuracy;

[0081] (2) Multi-fold cross-validation was performed on the accuracy assessment of the early warning to obtain stability indicators, and sensitivity analysis was conducted on the stability indicators to obtain key influencing factors;

[0082] (3) Construct a causal inference model for key influencing factors to obtain a potential causal relationship network, and conduct path analysis on the potential causal relationship network to obtain the risk propagation chain;

[0083] (4) Conduct scenario simulations of the risk propagation chain to obtain multiple possible risk evolution paths, and conduct probability assessments of multiple possible risk evolution paths to obtain risk level classifications;

[0084] (5) Perform multi-dimensional comprehensive scoring on risk level classification to obtain a risk severity matrix, and perform a structured description of the risk severity matrix to obtain a graded risk assessment report.

[0085] Specifically, similarity algorithms, such as cosine similarity or Euclidean distance, are used to compare the current warning situation with cases in the historical database, filtering out cases with similarity higher than a preset threshold to form a set of similar warning scenarios. Then, feedback analysis is performed on these similar warning scenario sets, comparing the warning triggering conditions with actual security events, and calculating the accuracy, recall, and F1 score of the warnings to obtain an assessment of warning accuracy. To ensure the reliability of the assessment results, multi-fold cross-validation is used for the warning accuracy assessment. Specifically, the dataset is divided into K subsets (usually K=5 or 10), and K-1 subsets are used as the training set each time, with the remaining subset used as the test set, repeated K times to obtain K evaluation results. The average of these results serves as the final stability index, reflecting the performance stability of the warning system under different data distributions. Next, sensitivity analysis is performed on the stability index, observing changes in warning performance by changing the values ​​of different parameters (such as threshold, time window size, etc.) to identify the key factors that have the greatest impact on warning results.

[0086] The first step involves constructing causal inference models for the identified key influencing factors. This step aims to reveal the intrinsic connections between risk factors. Commonly used causal inference methods include structural equation modeling (SEM) or Bayesian networks, which can handle complex multivariate relationships and consider potential confounding factors. These models yield a network of potential causal relationships describing the interactions between various risk factors. Subsequently, path analysis is performed on this network to calculate direct and indirect effects, identifying propagation paths between risk factors and thus obtaining risk propagation chains. Scenario simulations are then conducted on the obtained risk propagation chains. This step uses methods such as Monte Carlo simulation or agent-based modeling to simulate the evolution of risk under different initial conditions. Through multiple simulations, various possible risk evolution paths are obtained, reflecting the development trends and possible outcomes of risks under different circumstances. Probability assessments are performed on these risk evolution paths, calculating the probability of occurrence for each path. Based on the combination of risk level and probability of occurrence, risks are classified into different levels, such as low risk, medium risk, and high risk, thus obtaining a risk level classification.

[0087] Finally, a multi-dimensional comprehensive scoring is performed on the risk level classification. This process considers multiple factors, such as the probability of risk occurrence, the potential degree of loss, and the scope of impact, using methods such as weighted summation or fuzzy comprehensive evaluation to calculate a comprehensive score for each risk. These scores constitute a risk severity matrix, where each element represents the severity of a specific risk type under specific conditions. This risk severity matrix is ​​then structured, transforming the numerical results into easily understandable textual descriptions, including risk type, level, potential impact, and recommended measures, ultimately forming a graded and categorized risk assessment report. For example, in a large steel structure building project, the early warning system set an adaptive early warning trigger matrix based on 10 key indicators (such as main beam stress, node deformation, and ambient temperature). By matching with 200 historical cases from the past 5 years, 50 cases with a similarity greater than 0.8 were selected as a similar early warning scenario set. Feedback analysis showed that the system's early warning accuracy was 85%, recall was 90%, and F1 score was 0.87. Stability assessment of these results using 5-fold cross-validation yielded an average F1 score of 0.86 and a standard deviation of 0.03, indicating relatively stable system performance. Sensitivity analysis revealed that the ambient temperature threshold and the rate of change of main beam stress are the two most critical factors affecting the accuracy of the early warning system.

[0088] Based on these key factors, a Bayesian network with 15 nodes was constructed as a causal inference model. Path analysis revealed a major risk propagation chain: increased ambient temperature → thermal expansion of steel → increased node stress → intensified deformation of the main beam → decreased overall structural stability.

[0089] Through 1000 Monte Carlo simulations, 100 different risk evolution paths were generated. Probability assessments showed that 20% of the paths were high-risk, 30% medium-risk, and 50% low-risk. A multi-dimensional comprehensive scoring system considered three aspects: risk probability (weight 0.3), potential economic loss (weight 0.4), and safety impact (weight 0.3). The scoring results formed a 10×3 risk severity matrix (10 risk types, 3 severity levels). The final risk assessment report details the characteristics, possible consequences, and recommended countermeasures for each risk, providing the project management team with comprehensive risk management guidance.

[0090] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0091] (1) Semantic parsing of the risk assessment report of the classification and grading is performed to obtain key risk descriptors, and scenario elements are extracted from the key risk descriptors to obtain risk scenario models;

[0092] (2) Perform resource constraint analysis on the risk scenario model to obtain a list of available emergency resources, and prioritize the list of available emergency resources to obtain a resource allocation strategy;

[0093] (3) Decompose the resource allocation strategy into tasks to obtain a preliminary action step sequence, and optimize the timing logic of the preliminary action step sequence to obtain the optimized action flow.

[0094] (4) Assign roles to the optimized action flow to obtain a responsibility matrix, and design communication paths for the responsibility matrix to obtain an information transmission network;

[0095] (5) Identify decision points in the information transmission network to obtain a set of key decision nodes, and generate alternative solutions for the set of key decision nodes to obtain an executable multi-step emergency plan.

[0096] Specifically, natural language processing techniques, such as Named Entity Recognition (NER) and keyword extraction, are used to identify key risk descriptors from the risk assessment report. These descriptors may include information such as risk type, severity, and scope of impact. Subsequently, scenario elements are extracted from these key risk descriptors. Using knowledge graphs or ontology models, the abstract risk descriptions are transformed into concrete scenario elements, such as affected structural components, hazardous areas, and potential cascading effects, thereby constructing a structured risk scenario model. The next crucial step is to perform resource constraint analysis on the risk scenario model. This process first generates a list of currently available emergency resources, including manpower, equipment, and materials, based on the BIM model and resource management system. Then, multi-criteria decision analysis (MCDA) methods, such as the Analytic Hierarchy Process (AHP) or TOPSIS, are used to prioritize these resources. The prioritization is based on factors such as resource effectiveness, availability, and cost. The prioritization can be expressed by the following formula:

[0097]

[0098] Among them, P i It is the priority score of resource i, w j The weight of the j-th criterion, s ij Let be the score of resource i on standard j, and n be the number of evaluation standards. This yields an optimized resource allocation strategy. The resource allocation strategy is then decomposed into tasks using a Work Breakdown Structure (WBS) technique, breaking down the overall emergency response into a series of specific action steps. These initial action step sequences are then optimized using a time-series logic optimization method, employing Critical Path Method (CPM) or Process Review Technique (PERT), to determine the dependencies between steps and the optimal execution order. The optimized action flow can be represented by a network diagram, where nodes represent action steps and edges represent dependencies between steps. Time estimation can be performed using the following PERT formula:

[0099]

[0100] Among them, T e It is the expected time, T o This is the most optimistic time estimate, T m Most likely time estimate, T pThis is the most pessimistic time estimate. Roles are assigned to the optimized action flow using the Responsibility Assignment Matrix (RACI) technique. RACI represents four roles: Responsible, Accountable, Consulted, and Informed. By filling the RACI matrix, the executors and stakeholders of each action step are clearly defined, resulting in a detailed responsibility matrix. Based on this responsibility matrix, communication paths are designed, and graph theory algorithms such as Dijkstra's shortest path algorithm are used to optimize information flow, ultimately forming an efficient information transmission network.

[0101] Finally, decision points are identified within the information transmission network using decision tree analysis or influence graph techniques to pinpoint nodes requiring critical judgments. These nodes constitute the set of critical decision nodes. For each critical decision node, scenario planning techniques are used to generate multiple alternative solutions. The generation of these alternative solutions considers different risk development scenarios and resource availability, ultimately forming a comprehensive, actionable, multi-step emergency response plan.

[0102] For example, in a large steel structure building project, the risk assessment report identified "excessive deformation of the main beam" as a key risk descriptor through semantic parsing. Scene element extraction transformed this into a specific risk scenario model, including deformation location, degree, and potential impact range. Resource constraint analysis showed that currently available emergency resources included 5 hydraulic jacks, 3 sets of temporary support frames, and 2 emergency repair teams. Using the AHP method to prioritize these resources, the resulting resource allocation strategy was to first deploy temporary support frames, then use hydraulic jacks for correction, and finally have the repair teams reinforce the structure. Task decomposition broke down the entire emergency response into 10 specific steps, including site survey, safe zone delineation, and temporary support frame installation. The PERT method was used to optimize the execution order of these steps, calculating the expected completion time for the entire emergency response to be 6 hours, with the critical path including 5 key steps. Responsibility allocation used a RACI matrix to clearly define the responsible person for each step, such as the project manager for overall coordination, the safety supervisor for on-site safety management, and the engineer for technical guidance.

[0103] The information delivery network was designed to ensure that critical information could reach relevant personnel within an average of one minute. Decision point analysis identified three key decision nodes, including whether evacuation was necessary, which reinforcement method to adopt, and whether work stoppage was required. For these decision points, a total of 12 alternative plans were generated, forming the final multi-step emergency response plan.

[0104] The above describes the BIM-assisted steel structure construction safety management early warning method in the embodiments of this application. The following describes the BIM-assisted steel structure construction safety management early warning device in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the BIM-assisted steel structure construction safety management early warning device in this application includes:

[0105] Processing module 201 is used to collect multi-source heterogeneous data and perform spatiotemporal alignment processing on the steel structure construction site to obtain a high-dimensional synchronized engineering data stream;

[0106] Extraction module 202 is used to perform dynamic noise reduction and feature extraction on the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector;

[0107] Mining module 203 is used to perform time-series correlation analysis and pattern mining on the standardized multimodal feature vectors to obtain a set of potential risk indicators;

[0108] Evaluation module 204 is used to perform multi-level threshold calculation and context-sensitive evaluation on the potential risk indicator set to obtain an adaptive early warning trigger matrix;

[0109] Analysis module 205 is used to perform multi-dimensional cross-validation and causal chain analysis on the adaptive early warning trigger matrix to obtain a graded and classified risk assessment report;

[0110] The generation module 206 is used to perform scenario-based analysis and response plan generation on the risk assessment report of the hierarchical classification, so as to obtain an executable multi-step emergency plan.

[0111] Through the collaborative efforts of the aforementioned components, multi-source heterogeneous data acquisition and spatiotemporal alignment processing of the steel structure construction site yields a high-dimensional synchronized engineering data stream. This enables comprehensive and multi-faceted real-time monitoring of the construction site, providing a complete and accurate data foundation for subsequent analysis. Secondly, dynamic noise reduction and feature extraction are performed on the high-dimensional synchronized engineering data stream to obtain standardized multimodal feature vectors. This effectively eliminates noise interference in the data, extracts the most representative features, and improves the accuracy and efficiency of subsequent analysis. Thirdly, temporal correlation analysis and pattern mining are conducted on the standardized multimodal feature vectors to obtain a set of potential risk indicators. This deeply uncovers the implicit time dependencies and risk patterns within the data, providing a scientific basis for risk warning. Furthermore, multi-level threshold calculations and context-sensitive assessments are performed on the potential risk indicator set to obtain an adaptive warning trigger matrix. This enables dynamic adjustment of the warning mechanism, significantly improving the accuracy and adaptability of warnings. Simultaneously, multi-dimensional cross-validation and causal chain analysis of the adaptive warning trigger matrix yields a graded and categorized risk assessment report. This not only quantifies the degree of risk but also reveals the causal relationships between risk factors, providing comprehensive support for risk management decisions. Finally, the risk assessment reports, categorized by level and type, are analyzed in a scenario-based manner to generate response plans, resulting in an executable multi-step emergency plan. This achieves intelligent management of the entire process from risk identification to emergency response. This method greatly improves the accuracy, real-time performance, and operability of steel structure construction safety management, effectively reduces the probability of safety accidents, and enhances the efficiency and effectiveness of emergency response.

[0112] Based on the same technical concept, embodiments of this application also provide an electronic device. (Refer to...) Figure 3 The diagram shown is a structural schematic of an electronic device 300 provided in an embodiment of this application, including a processor 301, a memory 302, and a bus 303. The memory 302 is used to store execution instructions and includes a main memory 3021 and an external memory 3022. The main memory 3021, also called internal memory, is used to temporarily store computational data in the processor 301 and data exchanged with external memory 3022 such as a hard disk. The processor 301 exchanges data with the external memory 3022 through the main memory 3021. When the electronic device 300 is running, the processor 301 and the memory 302 communicate through the bus 303.

[0113] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the BIM-assisted steel structure construction safety management early warning method.

[0114] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0115] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A BIM-assisted method for safety management and early warning in steel structure construction, characterized in that, The BIM-assisted early warning method for safety management of steel structure construction includes: Multi-source heterogeneous data acquisition and spatiotemporal alignment processing are performed on the steel structure construction site to obtain a high-dimensional synchronized engineering data stream. Dynamic noise reduction and feature extraction are performed on the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector; The standardized multimodal feature vectors are subjected to time-series correlation analysis and pattern mining to obtain a set of potential risk indicators. This includes: performing time-series decomposition on the standardized multimodal feature vectors to obtain trend, periodic, and random components; performing autocorrelation analysis on the trend, periodic, and random components to obtain time-dependent characteristics; performing dynamic time warping on the time-dependent characteristics to obtain aligned feature sequences; performing sliding window segmentation on the aligned feature sequences to obtain local feature fragments; performing frequent pattern mining on the local feature fragments to obtain recurring feature combinations; performing association rule learning on the recurring feature combinations to obtain association rules between features; performing graph structure modeling on the association rules between features to obtain a feature association network; performing community detection on the feature association network to obtain highly correlated feature clusters; and performing importance scoring on the highly correlated feature clusters to obtain key feature subsets; and performing semantic interpretation on the key feature subsets to obtain a set of potential risk indicators. Multi-level threshold calculation and context-sensitive evaluation are performed on the potential risk indicator set to obtain an adaptive early warning trigger matrix; Multi-dimensional cross-validation and causal chain analysis are performed on the adaptive early warning trigger matrix to obtain a graded and classified risk assessment report; The risk assessment reports of the classification and categorization are analyzed in a scenario-based manner and response plans are generated to obtain an executable multi-step emergency plan.

2. The BIM-assisted steel structure construction safety management and early warning method according to claim 1, characterized in that, The process of acquiring multi-source heterogeneous data and performing spatiotemporal alignment processing at the steel structure construction site to obtain a high-dimensional synchronized engineering data stream includes: Three-dimensional laser scanning is performed on the steel structure components to obtain high-precision point cloud data. The high-precision point cloud data is then subjected to noise filtering and sparsification to obtain structural geometric contour data. Feature extraction and parametric modeling are performed on the structural geometric contour data to obtain a steel structure BIM model, and attribute information is added to the steel structure BIM model to obtain a semantic BIM model; Strain sensors are arranged at key nodes of the steel structure to obtain real-time strain data. The real-time strain data is then denoised and outlier detected to obtain an effective strain dataset. Environmental monitoring equipment is deployed at the construction site to obtain multidimensional environmental parameter data. The multidimensional environmental parameter data is then standardized and interpolated to obtain a continuous environmental state sequence. The construction personnel and equipment are configured with a real-time positioning system to obtain dynamic location information, and the dynamic location information is then subjected to trajectory smoothing and predictive analysis to obtain spatiotemporal behavior pattern data. The semantic BIM model, effective strain dataset, continuous environmental state sequence and spatiotemporal behavior pattern data are timestamped to obtain a multi-source data synchronization matrix. The multi-source data synchronization matrix is ​​then spatially registered and scaled to obtain a high-dimensional synchronized engineering data stream.

3. The BIM-assisted steel structure construction safety management and early warning method according to claim 1, characterized in that, The process of dynamically denoising and extracting features from the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector includes: An adaptive wavelet transform is performed on the high-dimensional synchronized engineering data stream to obtain a multi-scale decomposition result, and the multi-scale decomposition result is subjected to threshold denoising to obtain a smoothed data stream. Principal component analysis is performed on the smoothed data stream to obtain a dimensionality-reduced feature set, and nonlinear mapping is performed on the dimensionality-reduced feature set to obtain a higher-order feature representation; The higher-order feature representation is transformed in the time-frequency domain to obtain a time-frequency feature matrix, and peak detection and spectral analysis are performed on the time-frequency feature matrix to obtain key frequency features; Statistical moments are extracted from the key frequency features to obtain statistical feature vectors, and the statistical feature vectors are normalized to obtain a standardized feature set. Multimodal fusion is performed on the standardized feature set to obtain a fused feature tensor, and tensor decomposition is performed on the fused feature tensor to obtain a standardized multimodal feature vector.

4. The BIM-assisted steel structure construction safety management and early warning method according to claim 1, characterized in that, The process of performing multi-level threshold calculations and context-sensitive assessments on the potential risk indicator set to obtain an adaptive early warning trigger matrix includes: Historical data statistical analysis is performed on the potential risk indicator set to obtain the indicator distribution characteristics, and multi-level quantile calculation is performed on the indicator distribution characteristics to obtain the initial threshold set; The initial threshold set is fused with expert knowledge to obtain an adjusted threshold range, and the adjusted threshold range is then divided into fuzzy sets to obtain multi-level fuzzy thresholds. A dynamic time window is applied to the multi-level fuzzy thresholds to obtain a time-varying threshold sequence, and context information is associated with the time-varying threshold sequence to obtain a context-sensitive threshold. The context-sensitive thresholds are combined in multiple dimensions to obtain a threshold decision tree, and the threshold decision tree is pruned and optimized to obtain simplified decision rules. The simplified decision-making rules are matrix mapped to obtain the early warning triggering condition matrix, and an adaptive adjustment mechanism is designed for the early warning triggering condition matrix to obtain the adaptive early warning triggering matrix.

5. The BIM-assisted steel structure construction safety management and early warning method according to claim 1, characterized in that, The adaptive early warning trigger matrix is ​​subjected to multi-dimensional cross-validation and causal chain analysis to obtain a graded and classified risk assessment report, including: Historical case matching is performed on the adaptive early warning trigger matrix to obtain a set of similar early warning scenarios, and the results of the similar early warning scenario set are analyzed to obtain an early warning accuracy assessment. Multi-fold cross-validation was performed on the accuracy assessment of the early warning to obtain a stability index, and sensitivity analysis was conducted on the stability index to identify key influencing factors. A causal inference model is constructed for the key influencing factors to obtain a potential causal relationship network, and path analysis is performed on the potential causal relationship network to obtain the risk propagation chain; Scenario simulation is performed on the risk propagation chain to obtain multiple possible risk evolution paths, and probability assessment is performed on the multiple possible risk evolution paths to obtain risk level classification; The risk level classification is comprehensively scored from multiple dimensions to obtain a risk severity matrix, and the risk severity matrix is ​​described in a structured manner to obtain a graded risk assessment report.

6. The BIM-assisted early warning method for safety management of steel structure construction according to claim 1, characterized in that, The risk assessment report, classified and categorized, is analyzed in a scenario-based manner, and a response plan is generated to obtain an executable multi-step emergency plan, including: The risk assessment report of the graded classification is semantically parsed to obtain key risk descriptors, and the key risk descriptors are used to extract scenario elements to obtain a risk scenario model. Resource constraint analysis is performed on the risk scenario model to obtain a list of available emergency resources, and the list of available emergency resources is prioritized to obtain a resource allocation strategy. The resource allocation strategy is decomposed into tasks to obtain a preliminary action step sequence, and the preliminary action step sequence is optimized by time-series logic to obtain an optimized action flow. Roles are assigned to the optimized action flow to obtain a responsibility matrix, and communication paths are designed for the responsibility matrix to obtain an information transmission network; The information transmission network is used to identify decision points to obtain a set of key decision nodes, and alternative solutions are generated from the set of key decision nodes to obtain an executable multi-step emergency plan.

7. A BIM-assisted steel structure construction safety management early warning device, used to implement the BIM-assisted steel structure construction safety management early warning method as described in any one of claims 1 to 6, characterized in that, The BIM-assisted steel structure construction safety management early warning device includes: The processing module is used to collect and spatiotemporally align multi-source heterogeneous data from the steel structure construction site to obtain a high-dimensional synchronized engineering data stream. The extraction module is used to perform dynamic noise reduction and feature extraction on the high-dimensional synchronized engineering data stream to obtain a standardized multimodal feature vector; The mining module is used to perform temporal correlation analysis and pattern mining on the standardized multimodal feature vectors to obtain a set of potential risk indicators. This includes: performing time series decomposition on the standardized multimodal feature vectors to obtain trend, periodic, and random components; performing autocorrelation analysis on the trend, periodic, and random components to obtain time-dependent characteristics; performing dynamic time warping on the time-dependent characteristics to obtain aligned feature sequences; performing sliding window segmentation on the aligned feature sequences to obtain local feature fragments; performing frequent pattern mining on the local feature fragments to obtain recurring feature combinations; performing association rule learning on the recurring feature combinations to obtain association rules between features; performing graph structure modeling on the association rules between features to obtain a feature association network; performing community detection on the feature association network to obtain highly correlated feature clusters; and performing importance scoring on the highly correlated feature clusters to obtain a key feature subset; and performing semantic interpretation on the key feature subset to obtain a set of potential risk indicators. The evaluation module is used to perform multi-level threshold calculation and context-sensitive evaluation on the potential risk indicator set to obtain an adaptive early warning trigger matrix; The analysis module is used to perform multi-dimensional cross-validation and causal chain analysis on the adaptive early warning trigger matrix to obtain a graded and classified risk assessment report; The generation module is used to perform scenario-based analysis and response plan generation on the risk assessment reports of the hierarchical classification, so as to obtain an executable multi-step emergency plan.

8. A computer device, characterized in that, include: The system includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the BIM-assisted steel structure construction safety management and early warning method as described in any one of claims 1 to 6.

Images