Intelligent construction site safety evaluation method and system based on data elements

Abstract

The invention discloses an intelligent construction site safety evaluation method and system based on data elements, and relates to the technical field of computer platform load balancing, and the method comprises the steps: collecting construction site multi-source heterogeneous data, and carrying out the preprocessing of the construction site multi-source heterogeneous data; constructing a safety evaluation model of time sequence modeling and graph structure fusion through combination of a long short-term memory network and a graph neural network, performing feature selection and dimension compression operation on the processed multi-source heterogeneous data of the construction site to obtain an input feature tensor, and inputting the input feature tensor into the safety evaluation model; and calculating a comprehensive score according to a model output result, and generating a classification result and data feedback through an edge node. According to the intelligent construction site safety evaluation method based on the data elements provided by the invention, combined modeling of time sequence dependence and space association is realized through a deep modeling structure fusing LSTM and GNN, and the risk identification precision under multi-source heterogeneous data is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of informatization and intelligent safety management in construction engineering, and specifically to a safety evaluation method and system for intelligent construction sites based on data elements. Background Art

[0002] With the continuous improvement of the informatization level in the construction industry, as an important part of digital construction, intelligent construction sites have been widely applied in large-scale infrastructure, urban renewal and other projects. Relying on technologies such as the Internet of Things, big data, and artificial intelligence, intelligent construction sites achieve real-time perception, dynamic scheduling, and data-driven management of the construction site. Especially in terms of safety management, through the deployment of various sensors, video monitoring devices, and behavior recording terminals, it is possible to achieve full-cycle monitoring of construction workers, construction machinery, and the on-site environment.

[0003] The existing intelligent construction site safety evaluation system still has the problem that most current safety evaluation methods rely on static indicators or manual inspection data and fail to effectively integrate multi-source heterogeneous data such as video images, text logs, and sensor signals, resulting in limited evaluation dimensions and a single data structure. Most safety evaluation models use rule-driven or single algorithms for scoring, lacking the ability to model the complex correlations and dynamic change characteristics between data, and it is difficult to adapt to the fusion requirements of high-frequency, non-linear, and unstructured data. Some systems only perform risk judgments in a fixed threshold manner, lacking an adaptive adjustment mechanism for different construction stages, scenario environments, and risk types, which is prone to false alarms or missed alarms. At the model deployment level, most are concentrated on cloud processing, with insufficient response timeliness and unable to meet the requirements of local rapid response and dynamic adjustment for major safety hazards.

[0004] Therefore, there is an urgent need for an intelligent safety evaluation method and system for intelligent construction sites that can integrate multi-source heterogeneous data, have the ability of adaptive feature extraction and risk modeling, and support edge deployment and multi-task output, so as to improve the safety management level and risk response efficiency of construction sites. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is: the existing intelligent construction site safety evaluation methods have weak data fusion capabilities, low model generalization capabilities, slow risk identification and response, and how to construct a safety evaluation model with the ability of joint modeling of time series and structure based on multi-source heterogeneous data, and achieve efficient and interpretable intelligent risk classification and feedback.

[0007] To solve the above technical problems, the present invention provides the following technical solution: A safety evaluation method for intelligent construction sites based on data elements, including collecting multi-source heterogeneous data of the construction site.

[0008] Preprocess the multi-source heterogeneous data on the construction site.

[0009] Construct a safety evaluation model that integrates time series modeling and graph structure by combining long short-term memory network and graph neural network.

[0010] Perform feature selection and dimensionality compression operations on the processed multi-source heterogeneous data of the construction site to obtain an input feature tensor, and input the input feature tensor into the safety evaluation model.

[0011] Calculate the comprehensive score according to the model output result, and generate classification results and data feedback through edge nodes.

[0012] Performing feature selection and dimensionality compression operations to obtain an input feature tensor includes extracting semantic and structural features of text and images through BERT and ResNet respectively, and after dimensionality reduction by PCA, splicing them with structured features into a fused feature vector.

[0013] Use the Pearson correlation coefficient to filter redundant features, LASSO regression to achieve sparse selection, and random forest ranking to screen key variables to form a unified input feature tensor, which is input into the safety evaluation model integrating LSTM and GNN structures, and combines the channel attention mechanism to achieve multi-task prediction output.

[0014] The channel attention mechanism includes allocating dynamic weights according to the importance of different modality features under the task objective.

[0015] As a preferred solution of the intelligent construction site safety evaluation method based on data elements of the present invention, wherein: the collection of multi-source heterogeneous data on the construction site includes collecting personnel attendance data, video surveillance data, mechanical equipment operating status, environmental monitoring data, hydropower monitoring data, and safety inspection records as data sources.

[0016] Use a unified data access interface for protocol conversion, adopt a high-precision clock synchronization mechanism to timestamp various data streams, control the timestamp accuracy within seconds, and use an edge buffering mechanism in high-frequency data collection.

[0017] As a preferred solution of the intelligent construction site safety evaluation method based on data elements of the present invention, wherein: the preprocessing of the multi-source heterogeneous data on the construction site includes performing normalization, anomaly elimination, and time alignment processing on the collected multi-source heterogeneous data of the construction site.

[0018] The normalization process includes mapping the values of each dimension to the interval [0,1] using the min-max scaling method; Anomaly elimination includes processing based on the Z-score method, setting a deviation of three standard deviations as the threshold to mark and eliminate outliers.

[0019] The time alignment process includes mapping data from different sources to time windows based on the dynamic time warping algorithm, filling in missing time point data using the linear interpolation algorithm, and uniformly constructing time series structured data according to a preset sliding window structure.

[0020] As a preferred solution of the intelligent construction site safety evaluation method based on data elements according to the present invention, wherein: the construction of the safety evaluation model integrating time series modeling and graph structure includes using a long short-term memory network (LSTM) to process the normalized time series data and capture the time series dependencies of construction behaviors, equipment operations, and environmental fluctuations.

[0021] Based on the entities at the construction site, a graph neural network (GNN) model is constructed, with personnel, equipment, risk points, and environmental sources as the nodes of the graph, and the weights of the edges are set according to the historical co-occurrence probability and event interaction frequency between the nodes.

[0022] The outputs of the LSTM and GNN are respectively concatenated through a fusion layer, and the concatenated outputs of the LSTM and GNN are used as the inputs of the model prediction layer, and joint modeling and multi-task output are performed through the inputs of the model prediction layer.

[0023] The concatenation through the fusion layer includes introducing a feature channel attention mechanism and dynamically assigning weights to different modality information based on a task-specific loss function.

[0024] The multi-task output includes the predicted frequency of violation behaviors, the probability of abnormal states, and the risk coefficient vector.

[0025] The risk coefficient vector includes risk metrics for multiple dimensions such as the status of construction workers, abnormal environmental parameters, and abnormal equipment operations.

[0026] As a preferred solution of the intelligent construction site safety evaluation method based on data elements according to the present invention, wherein: obtaining the input feature tensor includes performing embedding processing on the text features and image features in the preprocessed multi-source heterogeneous data using BERT encoding and ResNet convolutional network, and extracting semantic vectors and structural graph features.

[0027] BERT encoding includes using a pre-trained language model based on the Transformer architecture to perform semantic vectorization processing on text data, and performing high-dimensional semantic modeling on natural language texts such as construction logs and safety inspection records through BERT encoding.

[0028] The ResNet convolutional network includes extracting features from image data such as video surveillance images, and alleviating the problem of gradient disappearance in the training of deep networks through a residual connection structure.

[0029] Perform principal component analysis dimensionality reduction on the feature subsets of text-based features and image-based features, and compress the embedding dimension to within 64 dimensions.

[0030] Principal component analysis includes extracting principal components by constructing a feature covariance matrix, retaining the most important variation directions between features, reducing dimensions and retaining information.

[0031] Unify and splice structured features, compressed semantic features, and image features to construct a fused feature vector, and perform feature screening operations based on the fused feature vector.

[0032] Performing feature screening operations includes calculating the Pearson correlation coefficient between each feature, identifying redundant feature pairs with an absolute correlation value greater than 0.8, and retaining one of them.

[0033] On the feature set after redundancy filtering, introduce the LASSO regression algorithm to construct a sparse feature selector with an L1 regularization term, and screen out features that do not contribute significantly to the model response variable.

[0034] Apply the random forest algorithm to the retained features to calculate the feature importance scores, and select the top 20% of the variables to form the final input feature tensor.

[0035] Take the feature set as the fused input feature tensor, input it into the safety evaluation model, and perform joint modeling and classification prediction.

[0036] The input feature tensor includes a multi-modal feature set after preprocessing, embedding encoding, dimensionality reduction compression, and screening, and a high-dimensional structured input matrix formed after dimensionality unification and structure rearrangement.

[0037] Structured features include subsets represented in numerical and time series forms in multi-source data.

[0038] As a preferred solution of the intelligent construction site safety evaluation method based on data elements according to the present invention, wherein: the calculation of the comprehensive score includes weighted combination of the results of multi-task outputs, setting a comprehensive score calculation formula, and setting interval grade mapping according to the score results. The interval grade mapping is divided into three grades: A, B, and C.

[0039] Grade A indicates excellent status, Grade B indicates controllable risks, and Grade C indicates obvious safety hazards that need to be rectified.

[0040] As a preferred solution of the intelligent construction site safety evaluation method based on data elements according to the present invention, wherein: the generation of classification results and data feedback includes deploying the trained safety evaluation model to the construction site edge server after compression and conversion into a lightweight model. The edge server receives the sensor data stream and data batches, and performs real-time inference operations by calling the deployed model, with an inference cycle not exceeding 1 time per hour.

[0041] When any output classification level in the inference result is A level, write the score and summary information into the local log as long-term learning data for the model.

[0042] When any output classification level in the inference result is B level, make a record. When B level appears continuously for three cycles, the system determines it as a risk trend state and performs a feedback operation.

[0043] When any output classification level in the inference result is C level, trigger the data feedback mechanism, and send the current score level, risk dimension distribution, and data segment identifier to the remote intelligent construction site platform through the configured HTTP interface, and record them in the edge log database for subsequent auditing and model correction sampling.

[0044] Another object of the present invention is to provide an intelligent construction site safety evaluation system based on data elements, which can solve the problems of low heterogeneous data processing efficiency, high model response delay, and lack of risk trend prediction ability in the current intelligent construction site safety management technology through real-time inference and multi-dimensional feedback mechanisms of lightweight models deployed at the edge.

[0045] As a preferred solution of the intelligent construction site safety evaluation system based on data elements according to the present invention, wherein: it includes a data acquisition module, a data preprocessing module, a model construction module, a safety evaluation module, and a score generation and data feedback module.

[0046] The data acquisition module is used to acquire multi-source heterogeneous data of the construction site.

[0047] The data preprocessing module is used to preprocess the multi-source heterogeneous data of the construction site.

[0048] The model construction module is used to construct a safety evaluation model that combines long short-term memory network and graph neural network to integrate time series modeling and graph structure.

[0049] The safety evaluation module is used to perform feature selection and dimension compression operations on the processed multi-source heterogeneous data of the construction site to obtain an input feature tensor, and input the input feature tensor into the safety evaluation model.

[0050] The score generation and data feedback module is used to calculate a comprehensive score according to the model output result, and generate classification results and data feedback through the edge node.

[0051] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of a smart construction site safety evaluation method based on data elements.

[0052] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of a smart construction site safety evaluation method based on data elements.

[0053] Advantages of the present invention: The smart construction site safety evaluation method based on data elements provided by the present invention realizes the joint modeling of temporal dependence and spatial association through a deep modeling structure that integrates LSTM and GNN, effectively improving the risk identification accuracy under multi-source heterogeneous data; by introducing BERT and ResNet to perform deep semantic extraction on text and image information, and jointly optimizing the feature weight distribution with a channel attention mechanism, the adaptability of the model to complex on-site environments is improved; through edge deployment of a lightweight inference model and a hierarchical feedback mechanism, the performance of the system in terms of response speed and local disposal ability is improved. The present invention achieves better results in terms of data fusion efficiency, model prediction accuracy, and safety response timeliness. Description of the Drawings

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0055] Figure 1 It is the overall flowchart of the smart construction site safety evaluation method based on data elements provided by the first embodiment of the present invention. Detailed Embodiments

[0056] To make the above objects, features, and advantages of the present invention more obvious and understandable, the detailed embodiments of the present invention will be described below with reference to the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0057] Embodiment 1, referring to Figure 1 , which is an embodiment of the present invention, provides a smart construction site safety evaluation method based on data elements, including: S1: Collect multi-source heterogeneous data of the construction site.

[0058] Collect the attendance data of workers, video surveillance data, operating status of mechanical equipment, environmental monitoring data, hydropower monitoring data, and safety inspection records as data sources.

[0059] Use a unified data access interface for protocol conversion, adopt a high-precision clock synchronization mechanism to timestamp various data streams, control the timestamp accuracy within seconds, and use an edge buffering mechanism in high-frequency data collection.

[0060] S2: Preprocess the multi-source heterogeneous data on the construction site.

[0061] Perform normalization, anomaly rejection, and time alignment processing on the collected multi-source heterogeneous data on the construction site.

[0062] The normalization process includes mapping the numerical values of each dimension to the interval [0, 1] using the min-max scaling method.

[0063] A preferred solution for the normalization process is: ; where represents the th normalized eigenvalue, represents the th data value, represents the minimum value of the feature dimension in the sample set, represents the maximum value of the feature dimension in the sample set.

[0064] Anomaly rejection includes processing based on the Z-score method, setting a deviation of three standard deviations as the threshold to mark and reject outliers.

[0065] A preferred solution for processing based on the Z-score method is: ; where represents the Z-score of the th sample point, represents the sample mean, represents the sample standard deviation.

[0066] Set the absolute value of the Z-score of the th sample point not to exceed 3. When it exceeds 3, it is determined as abnormal data.

[0067] Time alignment processing includes performing time window mapping on data from different sources based on the dynamic time warping algorithm, filling in missing time point data using the linear interpolation algorithm, and uniformly constructing time series structure data according to the preset sliding window structure.

[0068] S3: Combine the long short - term memory network with the graph neural network to construct a safety evaluation model that integrates time - series modeling and graph structure.

[0069] Use the long short - term memory network (LSTM) to process the normalized time - series data, capturing the temporal dependencies of construction behaviors, equipment operations, and environmental fluctuations.

[0070] Based on the entities at the construction site, construct a graph neural network (GNN) model. Consider personnel, equipment, risk points, and environmental sources as the nodes of the graph, and set the edge weights according to the historical co - occurrence probability and event interaction frequency between the nodes.

[0071] The outputs of LSTM and GNN are concatenated through a fusion layer. The concatenated outputs of LSTM and GNN are used as the input of the model prediction layer, and joint modeling and multi - task output are performed through the input of the model prediction layer.

[0072] The concatenation through the fusion layer includes introducing a feature channel attention mechanism and dynamically assigning weights to different modal information based on the task - specific loss function.

[0073] ; Among them, represents the fused feature of the th channel after being weighted by the channel attention mechanism, represents the feature vector of the original th channel, represents the channel weight coefficient, represents the corresponding loss function value for the task, represents the total number of task or modal channels, represents the channel index.

[0074] It should be noted that: the number of channels is equal to the number of tasks.

[0075] The multi - task output includes the prediction frequency of violation behaviors, the probability of abnormal states, and the risk coefficient vector.

[0076] The risk coefficient vector includes risk metrics for multiple dimensions such as the status of construction workers, abnormal environmental parameters, and abnormal equipment operations.

[0077] S4: Perform feature selection and dimensionality reduction operations on the processed multi - source heterogeneous data of the construction site to obtain an input feature tensor, and input the input feature tensor into the safety evaluation model.

[0078] For the text - type features and image - type features in the pre - processed multi - source heterogeneous data, use BERT encoding and ResNet convolutional network for embedding processing to extract semantic vectors and structural graph features.

[0079] BERT encoding includes the method of semantically vectorizing text data using a pre-trained language model based on the Transformer architecture, and high-dimensional semantic modeling of natural language texts such as construction logs and safety inspection records is performed through BERT encoding.

[0080] The ResNet convolutional network includes extracting features from image-like data such as video surveillance footage, and the gradient vanishing problem in the training of deep networks is alleviated through the residual connection structure.

[0081] Perform principal component analysis dimensionality reduction operation on the feature subsets of text-like features and image-like features, and compress the embedding dimension to within 64 dimensions.

[0082] A preferred solution for principal component analysis dimensionality reduction is: ; Among them, represents the original high-dimensional feature matrix, represents the principal component transformation matrix obtained by PCA calculation, represents the feature representation after dimensionality reduction.

[0083] The feature representation after dimensionality reduction is obtained by eigen-decomposition of the covariance matrix.

[0084] Principal component analysis includes extracting principal components by constructing a feature covariance matrix, retaining the most important variation directions between features, reducing dimensions and retaining information.

[0085] Unify and splice structured-like features, compressed semantic features and image features to construct a fused feature vector, and perform feature screening operations based on the fused feature vector.

[0086] Performing feature screening operations includes calculating the Pearson correlation coefficient between each feature, identifying redundant feature pairs with an absolute correlation value greater than 0.8, and retaining one of them.

[0087] A preferred solution for calculating the Pearson correlation coefficient between each feature is: ; Among them, represents the feature and the feature correlation coefficient, represents the feature in the th sample after normalization of the corresponding feature value, , represent the means of their respective features, the numerator is the covariance term, and the denominator is the product of the standard deviations.

[0088] On the feature set after redundant filtering, the LASSO regression algorithm is introduced to construct a sparse feature selector with an L1 regularization term, and features that do not contribute significantly to the model response variable are screened out.

[0089] A preferred scheme for constructing a sparse feature selector with an L1 regularization term is: ; where, represents the input feature vector of the th sample, represents the regression coefficient of the th feature, represents the L1 regularization hyperparameter, represents the number of samples, represents the number of feature dimensions.

[0090] Apply the random forest algorithm to the retained features to calculate the feature importance scores, and select the top 20% of the variables to form the final input feature tensor.

[0091] A preferred scheme for calculating the feature importance scores by the random forest algorithm is: ; where, represents the complete feature set, represents the th feature, represents the rank of the th feature after sorting by importance, represents the total number of features, represents the set of the top 20% key features retained.

[0092] Take the feature set as the fused input feature tensor and input it into the safety evaluation model for joint modeling and classification prediction.

[0093] The input feature tensor includes a multi-modal feature set after preprocessing, embedding encoding, dimensionality reduction compression and screening, and a high-dimensional structured input matrix formed after dimensionality unification and structure rearrangement.

[0094] Structured class features include subsets represented in numerical and time series forms in multi-source data.

[0095] S5: Calculate the comprehensive score according to the model output result, and generate the classification result and data feedback through the edge node.

[0096] Perform weighted combination on the results of multi-task output, set the comprehensive score calculation formula, and set the interval level mapping according to the score result. The interval level mapping is divided into three levels: A, B, and C.

[0097] A preferred solution for calculating the comprehensive score is as follows: ; Among them, represents the final comprehensive score, represents the sub-score of personnel behavior, represents the sub-score of equipment operation, represents the sub-score of environmental status, , , represents the weight coefficient corresponding to each scoring item, , , The sum of is equal to 1.

[0098] A preferred solution for setting the interval grade mapping according to the scoring result is as follows: ; Among them, grade A indicates excellent status, grade B indicates controllable risks, and grade C indicates obvious potential safety hazards that need to be rectified.

[0099] After the trained safety evaluation model is compressed and converted into a lightweight model, it is deployed to the edge server of the construction site. The edge server receives the sensor data stream and data batches, and performs real-time inference operations by calling the deployed model. The inference cycle is no higher than once per hour.

[0100] When any output classification grade in the inference result is grade A, write the score and summary information into the local log as long-term learning data for the model.

[0101] When any output classification grade in the inference result is grade B, make a record. When grade B appears continuously for three cycles, the system determines the risk trend status and performs a feedback operation.

[0102] When any output classification grade in the inference result is grade C, trigger the data feedback mechanism, and send the current score grade, risk dimension distribution, and data segment identifier to the remote intelligent construction site platform through the configured HTTP interface, and record them in the edge log database for subsequent auditing and model correction sampling.

[0103] Example 2, which is an embodiment of the present invention, provides an intelligent construction site safety evaluation system based on data elements, including a data acquisition module, a data preprocessing module, a model construction module, a safety evaluation module, a score generation and data feedback module.

[0104] The data acquisition module is used to collect multi-source heterogeneous data of the construction site.

[0105] The data preprocessing module is used to preprocess the multi-source heterogeneous data of the construction site.

[0106] The model construction module is used to construct a safety evaluation model that combines time series modeling and graph structure fusion by integrating long short-term memory networks and graph neural networks.

[0107] The safety evaluation module is used to perform feature selection and dimensionality compression operations on the processed multi-source heterogeneous data of the construction site to obtain an input feature tensor, and input the input feature tensor into the safety evaluation model.

[0108] The generated score and data feedback module is used to calculate a comprehensive score based on the model output results, and generate classification results and data feedback through edge nodes.

Claims

1. A safety evaluation method for intelligent construction sites based on data elements, characterized in that, Including: Collecting multi-source heterogeneous data of the construction site; Preprocessing the multi-source heterogeneous data of the construction site; Constructing a safety evaluation model that integrates time series modeling and graph structure by combining long short-term memory network and graph neural network; Performing feature selection and dimensionality compression operations on the processed multi-source heterogeneous data of the construction site to obtain an input feature tensor, and inputting the input feature tensor into the safety evaluation model; Calculating a comprehensive score according to the model output result, and generating a classification result and data feedback through an edge node; Performing feature selection and dimensionality compression operations to obtain an input feature tensor, including extracting semantic and structural features of text and images through BERT and ResNet respectively, and splicing them with structured features into a fusion feature vector after PCA dimensionality reduction; Using Pearson correlation coefficient to filter redundant features, LASSO regression to achieve sparse selection, and random forest ranking to screen key variables to form a unified input feature tensor, which is input into the safety evaluation model integrating LSTM and GNN structures, and combining channel attention mechanism to achieve multi-task prediction output; The channel attention mechanism includes assigning dynamic weights according to the importance of different modal features under the task objective.

2. The intelligent construction site safety evaluation method based on data elements according to claim 1, wherein: The collecting multi-source heterogeneous data of the construction site includes Collecting personnel attendance data, video surveillance data, mechanical equipment operation status, environmental monitoring data, hydropower monitoring data, and safety inspection records as data sources; Using a unified data access interface for protocol conversion, adopting a high-precision clock synchronization mechanism to timestamp various data streams, controlling the timestamp accuracy within seconds, and using an edge buffering mechanism in high-frequency data collection.

3. The intelligent construction site safety evaluation method based on data elements as claimed in claim 2, wherein: The preprocessing the multi-source heterogeneous data of the construction site includes Performing normalization, anomaly rejection, and time alignment processing on the collected multi-source heterogeneous data of the construction site; The normalization processing includes mapping the numerical values of each dimension to the [0,1] interval using the min-max scaling method; Anomaly rejection includes processing based on the Z-score method, setting a deviation of three standard deviations as the threshold to mark and reject outliers; Time alignment processing includes performing time window mapping on data from different sources based on the dynamic time warping algorithm, filling in missing time point data using the linear interpolation algorithm, and uniformly constructing time series structure data according to the preset sliding window structure.

4. The intelligent construction site safety evaluation method based on data elements according to claim 3, wherein: The constructing a safety evaluation model that integrates time series modeling and graph structure includes Using the long short-term memory network LSTM to process the normalized time series data, capturing the time series dependence relationships of construction behaviors, equipment operations, and environmental fluctuations; Based on the entities at the construction site, constructing a graph neural network GNN model, taking personnel, equipment, risk points, and environmental sources as nodes of the graph, and setting the edge weights according to the historical co-occurrence probability and event interaction frequency between nodes; The outputs of LSTM and GNN are respectively spliced through a fusion layer, and the spliced outputs of LSTM and GNN are used as the input of the model prediction layer, and joint modeling and multi-task output are performed through the input of the model prediction layer; The splicing through the fusion layer includes introducing a feature channel attention mechanism and assigning dynamic weights to different modal information based on a task-specific loss function; The multi-task output includes the prediction frequency of violations, the probability of abnormal status, and the risk coefficient vector; The risk coefficient vector includes risk metrics for multiple dimensions such as the status of construction workers, abnormal environmental parameters, and abnormal equipment operation.

5. The intelligent construction site safety evaluation method based on data elements according to claim 4, wherein: The obtained input feature tensor includes Embedding processing of text features and image features in the preprocessed multi-source heterogeneous data using BERT encoding and ResNet convolutional network to extract semantic vectors and structural graph features; BERT encoding includes using a pre-trained language model based on the Transformer architecture to perform semantic vectorization on text data, and performing high-dimensional semantic modeling on natural language texts such as construction logs and safety inspection records through BERT encoding; The ResNet convolutional network includes extracting features from image data such as video surveillance footage and alleviating the problem of gradient disappearance in deep network training through a residual connection structure; Performing principal component analysis dimensionality reduction on the feature subsets of text features and image features, and compressing the embedding dimension to within 64 dimensions; Principal component analysis includes extracting principal components by constructing a feature covariance matrix, retaining the most important change directions between features, reducing dimensions, and retaining information; Unifying and splicing structured features, compressed semantic features, and image features to construct a fused feature vector, and performing feature screening operations based on the fused feature vector; Performing feature screening operations includes calculating the Pearson correlation coefficient between each feature, identifying redundant feature pairs with an absolute correlation value greater than 0.8, and retaining one of them; On the feature set after redundancy filtering, introduce the LASSO regression algorithm to construct a sparse feature selector with an L1 regularization term, and screen out features that do not contribute significantly to the model response variable; Applying the random forest algorithm to the retained features to calculate the feature importance scores, and selecting the top 20% of the variables to form the final input feature tensor; Taking the feature set as the fused input feature tensor and inputting it into the safety evaluation model for joint modeling and classification prediction; Input feature tensor Includes a multi-modal feature set that has been preprocessed, embedded encoded, dimensionally compressed, and screened, and a high-dimensional structured input matrix formed after dimension unification and structure rearrangement; Structured features include subsets represented in numerical and time series forms in multi-source data.

6. The intelligent construction site safety evaluation method based on data elements according to claim 5, wherein: The calculation of the comprehensive score includes Performing weighted combination on the results of multi-task output, setting the comprehensive score calculation formula, and setting interval level mapping according to the score results. The interval level mapping is divided into three levels: A, B, and C; Level A indicates excellent status, level B indicates controllable risk, and level C indicates obvious safety hazards that need to be rectified.

7. The intelligent construction site safety evaluation method based on data elements according to claim 6, wherein: The generation of classification results and data feedback includes Compressing and converting the trained safety evaluation model into a lightweight model and deploying it to the construction site edge server. The edge server receives sensor data streams and data batches, and performs real-time inference operations by calling the deployed model. The inference cycle is no higher than 1 time per hour; When any output classification level in the inference result is level A, write the score and summary information into the local log as long-term learning data for the model; When any output classification level in the inference result is level B, record it. When level B appears in three consecutive cycles, the system determines it as the risk trend status and performs a feedback operation; When any output classification level in the inference result is level C, trigger the data feedback mechanism. Send the current scoring level, risk dimension distribution, and data segment identifier to the remote intelligent construction site platform through the configured HTTP interface, and record them in the edge log database for subsequent auditing and model correction sampling.

8. A system adopting the intelligent construction site safety evaluation method based on data elements as described in any one of claims 1 to 7, characterized in that: It includes a data collection module, a data preprocessing module, a model construction module, a safety evaluation module, a scoring and data feedback generation module; The data collection module is used to collect multi-source heterogeneous data of the construction site; The data preprocessing module is used to preprocess the multi-source heterogeneous data of the construction site; The model construction module is used to construct a safety evaluation model that combines time series modeling and graph structure fusion through the combination of long short-term memory network and graph neural network; The safety evaluation module is used to perform feature selection and dimension compression operations on the processed multi-source heterogeneous data of the construction site to obtain an input feature tensor, and input the input feature tensor into the safety evaluation model; The scoring and data feedback generation module is used to calculate a comprehensive score based on the model output result, and generate a classification result and data feedback through the edge node.

9. A computer device, including a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for intelligent construction site safety evaluation based on data elements according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for intelligent construction site safety evaluation based on data elements according to any one of claims 1 to 7.

Images