An engineering project safety supervision system and method based on the Internet of Things

Abstract

This invention discloses an IoT-based engineering project safety supervision system and method, relating to the field of engineering management technology. The system includes: determining interaction behavior types based on the real-time location information of workers and the operating status of construction equipment, and classifying risk scenarios based on these interaction behavior types; extracting worker behavior features using a convolutional neural network fused with sparse perception based on the risk scenarios, and identifying abnormal behavior types based on the asymmetric association rules between the behavior features and the risk scenarios; automatically retrieving historical operating data of construction equipment based on the risk scenarios associated with the abnormal behavior types, and extracting abnormal equipment state features that trigger the abnormal behavior types; and locating potential hazards in the construction equipment based on the asymmetric association rules between the abnormal equipment state features and the abnormal behavior types. This invention achieves accurate identification of abnormal behaviors on-site and efficient location of potential equipment hazards, significantly improving the timeliness and accuracy of safety supervision.

Description

Technical Field

[0001] This invention relates to the field of engineering management technology, specifically to an Internet of Things-based engineering project safety monitoring system and method. Background Technology

[0002] As the scale and complexity of construction projects continue to increase, the interaction between on-site workers and large construction equipment is becoming more frequent, leading to a continuous increase in construction safety risks. Traditional construction safety supervision methods mainly rely on manual inspections by on-site management personnel and video monitoring, which suffer from limited monitoring coverage, insufficient real-time performance, and low accuracy in identifying abnormal behavior. Furthermore, while existing construction safety supervision methods based on IoT technology can achieve real-time data collection and preliminary analysis at the construction site, they still fall short in terms of refined identification of worker behavior characteristics and precise location of potential hazards in construction equipment.

[0003] In particular, current technologies for extracting personnel behavioral characteristics do not effectively address the sparsity of feature channel responses in construction scenarios, limiting the accuracy of abnormal behavior identification. Furthermore, the methods for analyzing the correlation between abnormal equipment status characteristics and abnormal personnel behavior are relatively simple, making it difficult to discover and effectively utilize the asymmetric relationships involved. This, in turn, affects the accurate location and timely warning of potential equipment hazards. Therefore, there is an urgent need for a project safety supervision method capable of accurately identifying abnormal interactions between personnel and equipment at construction sites, accurately locating potential equipment hazards, and sending safety alarms, in order to effectively improve the efficiency of safety supervision and accident prevention capabilities at construction sites. Summary of the Invention

[0004] The purpose of this invention is to provide an Internet of Things-based engineering project safety monitoring system and method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for safety supervision of engineering projects based on the Internet of Things, characterized in that it includes: The types of interactive behaviors are determined based on the real-time location information of the workers and the operating status of the construction equipment, and risk scenarios are classified according to the types of interactive behaviors. Based on risk scenarios, we use convolutional neural networks that integrate sparse perception to extract the behavioral features of workers, and identify abnormal behavior types based on the asymmetric association rules between behavioral features and risk scenarios. Once the abnormal behavior type is identified, the historical operating data of the construction equipment is automatically retrieved based on the risk scenario associated with the abnormal behavior type, and the abnormal state characteristics of the equipment that caused the abnormal behavior type are extracted from it. Based on the asymmetric association rules between abnormal equipment status characteristics and abnormal behavior types, the potential hazards of the construction equipment are located in reverse, and safety alarm information is pushed to the terminal equipment of the operators according to the potential hazards.

[0006] Secondly, the present invention provides an Internet of Things (IoT)-based engineering project safety monitoring system, implemented based on the aforementioned IoT-based engineering project safety monitoring method, comprising: The scenario segmentation module is used to determine the type of interaction behavior based on the real-time location information of the workers and the operating status of the construction equipment, and to classify risk scenarios according to the type of interaction behavior. The behavior recognition module is used to extract the behavioral features of workers based on risk scenarios using a convolutional neural network that integrates sparse perception, and to identify abnormal behavior types based on the asymmetric association rules between behavioral features and risk scenarios. The feature extraction module is used to automatically retrieve historical operating data of construction equipment based on the risk scenarios associated with the abnormal behavior type after the abnormal behavior type is identified, and extract the abnormal state features of the equipment that caused the abnormal behavior type from it. The safety management module is used to reverse locate the potential hazards of construction equipment based on the asymmetric association rules between the characteristics of abnormal equipment status and the types of abnormal behavior, and to push safety alarm information to the terminal equipment of the operators based on the potential hazards.

[0007] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention integrates a sparse-perception convolutional neural network to sparsify the behavioral characteristics of construction site workers, effectively highlighting key feature channels related to abnormal behavior and suppressing redundant information interference. This significantly improves the accuracy and robustness of identifying abnormal behavior types of workers, overcoming the problem of high misjudgment rates of abnormal behavior caused by traditional methods not paying sufficient attention to feature sparsity.

[0008] This invention avoids the drawbacks of interference from symmetrical rules in traditional association analysis by constructing asymmetric association rules between behavioral characteristics and risk scenarios. It effectively mines and utilizes the unidirectional and strongly indicative association between abnormal behavior and behavioral characteristics, achieving accurate identification of abnormal behavior types of construction site workers and reducing false alarm rates, thereby improving the pertinence of safety supervision measures.

[0009] This invention uses asymmetric association rules based on the abnormal state characteristics and abnormal behavior types of equipment to reverse locate potential hazards in construction equipment. This allows for precise identification of specific parts of construction equipment that may cause abnormal behavior, enabling targeted safety hazard investigation and alarm push. It effectively avoids the problem of inaccurate hazard location in traditional methods that leads to delayed or ineffective safety warnings, and significantly improves the response speed and accident prevention effect of on-site safety supervision. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0011] Figure 1 This is a flowchart of an IoT-based engineering project safety supervision method according to the present invention; Figure 2 This is a framework diagram of an Internet of Things-based engineering project safety monitoring system according to the present invention. Detailed Implementation

[0012] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make the description of this application more complete and comprehensive, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The drawings are merely illustrative illustrations of this application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0013] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of the exemplary embodiments disclosed in this application. However, those skilled in the art will recognize that the technical solutions disclosed in this application can be practiced with one or more specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the disclosure of this application.

[0014] Example 1, such as Figure 1 As shown, a method for safety supervision of engineering projects based on the Internet of Things is disclosed, including: S101: Determine the type of interaction behavior based on the real-time location information of the workers and the operating status of the construction equipment, and classify the risk scenarios according to the type of interaction behavior; In a specific implementation, the step of determining the interaction behavior type based on the real-time location information of the workers and the operating status of the construction equipment, and classifying risk scenarios based on the interaction behavior type, includes: Obtain the real-time relative position of the workers to the construction equipment; In this embodiment, the "real-time relative position of the operator to the construction equipment" refers to the real-time spatial position of the construction equipment and the operator monitored by a real-time positioning sensing device installed at the construction site, and the spatial distance and orientation relationship between the operator and the equipment calculated based on their position coordinates.

[0015] It should be further noted that the real-time positioning sensing device includes, but is not limited to: ultra-wideband (UWB) positioning device, radio frequency identification (RFID) device, Bluetooth beacon positioning device, GPS positioning device, visual camera positioning device or any combination thereof, preferably a UWB positioning device with high accuracy, used to collect the location data of workers and construction equipment in real time.

[0016] In practical implementation, taking a tower crane at a construction site as an example, UWB positioning tags are installed on the main structure of the tower crane and on the safety helmets or vests of the workers. The three-dimensional spatial coordinate data of the tower crane and the workers are obtained in real time through UWB base stations deployed on site, and then the real-time spatial distance between the two is calculated using the following formula: ; In the formula: Spatial distance between operators and equipment; : The real-time three-dimensional coordinates of the operator's location; : Real-time three-dimensional coordinates of the location of construction equipment (such as tower cranes).

[0017] Based on the trend of the duration of the construction equipment's operation, determine the current operating stage of the construction equipment; The "operating status of construction equipment" mentioned in this embodiment refers to operating parameter data that can reflect the current operating status of the equipment, including but not limited to: equipment start-up or stop status, current load size, operating frequency, power consumption, vibration characteristics and other equipment status data.

[0018] It should be further explained that this embodiment first collects the above-mentioned operating status parameters in real time through sensors (such as current sensors, power sensors, vibration sensors, etc.) pre-installed on the construction equipment; then, by analyzing the trend of the continuous duration of the equipment operating status parameters, the current operating stage of the equipment is determined. The operating stage specifically includes, but is not limited to: the initial start-up stage, the stable operation stage, the load change stage, and the preparation to stop stage.

[0019] For example, taking equipment power data as an example, the operating power data of the equipment is first monitored in real time through the equipment power sensor. Then, the trend analysis of the power data over a recent period (such as the last 5 minutes) is performed using the moving average method or exponential smoothing method. The specific calculation method is as follows: Moving average trend calculation method: ; In the formula: : The moving average of the power data at time t; : Power data at the i-th sampling time; N: Moving average window length (e.g., number of data points in the most recent minute).

[0020] Trend judgment criteria: If the moving average power continues to increase and tends to stabilize, it is determined to be the "initial stage of equipment startup"; If the power moving average value remains within a certain range for a long period of time, it is determined to be a "stable operation phase"; If the power moving average fluctuates significantly within a short period of time, it is determined to be a "load change phase"; If the power moving average shows a downward trend and the duration exceeds a preset threshold (e.g., a continuous decrease for 3 minutes), it is determined to be in the "preparation to stop phase".

[0021] Based on the combination of the relative positional relationship of the workers and the operational phase of the construction equipment, the types of interactive behaviors between the workers and the construction equipment are determined. Specifically, the "interaction behavior type" mentioned in this embodiment refers to the type of operation behavior determined by comprehensively considering the relative positional relationship between the operator and the equipment and the equipment operation stage, including but not limited to: non-contact observation behavior, equipment operation behavior, accidental entry into the equipment's dangerous area, equipment maintenance behavior, etc.

[0022] It should be further noted that this embodiment identifies and determines the type of interactive behavior by constructing a combined feature matrix of positional relationship and operational phase. The specific implementation method is as follows: For example, several behavior categories are first defined based on the operational phase of the construction equipment, such as: Equipment startup phase: generally defined as "Keep away". During the stable operation phase of the equipment: personnel are permitted to perform routine operations or equipment maintenance; During the equipment shutdown preparation phase: It is recommended that personnel evacuate or maintain a distance for observation.

[0023] At the same time, the safety level is defined based on the real-time spatial distance between the workers and the equipment: If the real-time distance is ≤3 meters, it is defined as "close range"; If the real-time distance is >3 meters and ≤5 meters, it is defined as "medium distance"; If the real-time distance is greater than 5 meters, it is defined as "long distance".

[0024] Based on the spatial distance and device operation stage mentioned above, a combined feature matrix is ​​constructed and the interaction behavior type is determined, as shown in Table 1 below: Table 1: Data Table of Interaction Behavior Types .

[0025] Furthermore, to achieve a reasonable determination of the hazardous area level of equipment based on the type of interaction behavior, this embodiment establishes a correspondence between the two through the following mapping relationship: When the interaction behavior type is "accidental entry into a dangerous area of ​​the device", the interaction behavior type is mapped to "high-risk area". When the interaction behavior type is "device operation behavior" or "device maintenance behavior", the interaction behavior type is mapped to "medium risk area". When the interaction behavior type is "non-contact observation behavior", the interaction behavior type is mapped to "low-risk area".

[0026] Based on the combination of equipment hazard areas and personnel exposure levels involved in the interaction behavior type, corresponding risk scenarios are classified; The “risk scenario” classification mentioned in this embodiment refers to the risk level determined by further cross-analyzing the equipment hazard area level and personnel exposure level in combination with the above-mentioned interaction behavior types, specifically including but not limited to high-risk scenarios, medium-risk scenarios, and low-risk scenarios.

[0027] It should be further noted that in this embodiment, the "personnel exposure level" is specifically quantified and determined by the time personnel spend in a specific area, for example: If the dwell time is greater than or equal to the preset high exposure threshold (e.g., 2 minutes), it is defined as "high exposure"; If the dwell time is between the medium exposure threshold (e.g., 30 seconds) and the high exposure threshold, it is defined as "medium exposure"; If the dwell time is less than the medium exposure threshold, it is defined as "low exposure".

[0028] A risk matrix is ​​generated by combining the equipment hazard level and exposure level, as shown in Table 2 below: Table 2: Risk Matrix Data Table .

[0029] S102: Based on risk scenarios, use a convolutional neural network that integrates sparse perception to extract the behavioral features of workers, and identify abnormal behavior types based on the asymmetric association rules between behavioral features and risk scenarios. In a specific implementation, the convolutional neural network fused with sparse awareness includes: The convolutional feature extraction module is used to extract local spatial features from worker behavior data; The sparse perception module, connected to the convolutional feature extraction module, is used to perform sparse masking processing on the convolutional features and output a sparse feature map. Specifically, the sparse sensing module includes: The sparse weight calculation unit is used to calculate the sparse weights of each feature channel based on the response intensity distribution of each channel in the convolutional feature. The feature channel selection unit is connected to the sparse weight calculation unit and is used to select high-weight feature channels according to the sparse weight. A sparse mask generation unit, connected to the feature channel selection unit, is used to generate a feature mask based on the selected feature channel. The sparse feature output unit, connected to the sparse mask generation unit, is used to apply a feature mask to the original convolutional features and output a sparse feature map.

[0030] The behavior feature aggregation module, connected to the sparse perception module, is used to generate a global behavior feature representation of the operator based on the sparse feature map; The behavior feature output layer, connected to the behavior feature aggregation module, is used to output the behavior feature vector of the operator fused with sparse perception.

[0031] Specifically, the convolutional feature extraction module is mainly used to extract spatial features from video image data of workers acquired on-site, in order to obtain initial behavioral features of local areas.

[0032] It should be further noted that the video image data in this embodiment is acquired through high-definition video cameras deployed at the construction site, with an image acquisition frequency of 30 frames per second and a resolution of 1920×1080 pixels. Each frame of video image acquired in real time is used as input data, with an input data size of H×W×C, where H is the image height (e.g., 1080 pixels), W is the image width (e.g., 1920 pixels), and C is the number of image channels (e.g., RGB three channels).

[0033] In specific implementation, the convolutional feature extraction module of this embodiment includes multiple convolutional units. Each convolutional unit consists of a convolutional layer, a batch normalization (BN) layer, and an activation function (such as ReLU). The specific processing procedure is as follows: First, the input video image is convolved using a convolutional layer to extract local spatial features. The calculation formula for the convolutional layer is as follows: ; In the formula, The output feature map after convolution is located at... ,aisle eigenvalues; Input image data at location The pixel value of channel c; : Convolution kernel in channel c and The weight of the positions between them; aisle The bias term.

[0034] Secondly, after the convolutional layers, the output features of the convolutions are standardized using batch normalization layers to improve the stability of network training. The specific calculation formula is as follows: ; In the formula: This represents the mean of the feature maps in the current batch; This represents the variance of the feature maps in the current batch; For training parameters of the batch normalized layer; It is a minimum value (e.g., 1e-5) used to prevent the denominator from being zero.

[0035] The standardized features are then non-linearly activated using an activation function layer (ReLU), calculated as follows: ; After processing by the above convolutional units, a convolutional feature map reflecting the local spatial characteristics of the workers' behavior is initially obtained.

[0036] Furthermore, the sparse perception module is used to perform sparsification processing on the initial convolutional feature map output by the convolutional feature extraction module, thereby highlighting key feature channels and suppressing redundant feature channels to enhance the distinguishability of behavioral features. Specifically, it includes: a sparse weight calculation unit, a feature channel selection unit, a sparse mask generation unit, and a sparse feature output unit.

[0037] Specifically, the implementation of the sparse weight calculation unit is as follows: First, global average pooling is performed on each channel feature map of the initial input convolutional feature map to obtain the global response intensity of each channel feature map: ; In the formula, Let be the global average response intensity of the c-th channel; For the c-th channel, position eigenvalues; These represent the height and width of the feature map, respectively.

[0038] Secondly, based on the obtained global response intensity of each channel, the sparse weight of each channel is calculated using the following formula: ; In the formula, is the sparse weight of the c-th channel; C is the total number of channels in the feature map.

[0039] Subsequently, the feature channel selection unit selects feature channels according to the aforementioned sparse weights, specifically as follows: Set a sparse weight threshold (e.g., 0.05). Channels with sparse weights greater than this threshold are selected as the set of valid feature channels, as follows: ; In the formula, The selected set of valid channels; The sparse weight threshold is (e.g., 0.05).

[0040] Then, the sparse mask generation unit generates a sparse mask based on the selected set of effective feature channels. The mask is specifically as follows: ; Finally, the sparse feature output unit applies the above sparse mask to the initial convolutional feature map to obtain a sparse feature map: ; Through the above implementation steps, a sparse feature map containing key feature channels is obtained for subsequent behavior feature aggregation and abnormal behavior type identification.

[0041] In a specific implementation, the behavior feature aggregation module is used to further process the sparse feature map output by the sparse perception module to obtain a global feature representation that can comprehensively characterize the behavior features of the workers. The specific implementation steps are as follows: First, in this embodiment, the behavior feature aggregation module performs a global pooling operation on the sparse feature map to obtain a global behavior feature vector. Specifically, this embodiment preferably uses Global Average Pooling (GAP). The specific calculation formula is as follows: ; In the formula: : The feature value of the c-th channel after global average pooling; : Sparse feature map at location The eigenvalue of the c-th channel; The spatial height and width of the feature map.

[0042] After the above pooling operation, the vector composed of the feature values ​​of each channel is used as the global behavior feature vector of the operator: ; Furthermore, the behavior feature output layer is specifically used to map the obtained global behavior feature vectors to a higher-dimensional feature space with stronger expressive power, for accurate classification and identification of abnormal behavior types. The specific implementation is as follows: The behavioral feature output layer described in this embodiment includes at least one fully connected layer (FC layer) and a corresponding activation function. Specifically, the behavioral feature output layer is calculated as follows: First, a fully connected layer is used to linearly map the global behavioral feature vector, as shown in the following formula: ; In the formula: : The feature vector after linear mapping; : Weight matrix of fully connected layer; : Bias vector of fully connected layer.

[0043] Secondly, the above feature vectors are nonlinearly mapped using a nonlinear activation function (such as the ReLU function) to obtain the final worker behavior feature vectors: ; in, : A feature vector of worker behavior fused with sparse perception.

[0044] In a specific implementation, the method for obtaining the sparse-aware convolutional neural network includes: Construct training samples of worker behavior characteristics and extract preliminary convolutional features using an initial convolutional neural network model; Based on the sparse distribution of channel responses in the initial convolutional features, a sparse target mask is generated for constrained training. A sparse target mask is introduced into the model training loss function, and the network weights are updated based on the backpropagation algorithm. Specifically, this embodiment implements the use of sparse target masks in the training process of convolutional neural networks in the following ways: First, the sparse target mask is introduced into the loss function used for network training as a regularization constraint. Specifically, a sparse regularization loss function is defined. as follows: ; In the formula, The actual response intensity of the c-th channel, (c) is the binary value (0 or 1) of the sparse target mask in the c-th channel.

[0045] Furthermore, the aforementioned sparse regularization loss is combined with the original classification task loss function (e.g., cross-entropy loss) to form a new overall training loss function. : ; In the formula, 𝜆 is the regularization coefficient used to balance the classification task and the sparse constraint, and the specific value is determined through experiments (e.g., between 0.1 and 1.0).

[0046] During network training, this embodiment utilizes the backpropagation algorithm, based on the loss function. The network weights are updated using gradients, and the specific weight update formula is as follows: ; in, For learning rate, This represents the weight parameters of the network.

[0047] Repeat the training iterations until the sparsity of the behavioral features output by the network reaches the requirements of the sparse target mask, and obtain a convolutional neural network that integrates sparse perception. To determine the convergence of network training, this embodiment defines the feature sparsity evaluation metric SparseRatio as: ,in, C represents the number of channels in the current network output features whose response intensity exceeds the sparsity threshold (e.g., 0.05), and C represents the total number of channels in the feature map.

[0048] During training, this embodiment calculates the above-mentioned sparsity evaluation metric SparseRatio after each training iteration. When the difference between SparseRatio and the effective channel ratio in the sparse target mask is less than a preset convergence threshold (e.g., 0.01), the network training is considered to have converged, and the training ends, resulting in a convolutional neural network with sparse awareness.

[0049] In a specific implementation, the step of identifying abnormal behavior types based on the asymmetric association rules between behavioral characteristics and risk scenarios includes: Obtain the correlation frequency between worker behavior characteristics and abnormal behavior types in different risk scenarios; In specific implementation, this embodiment first analyzes the behavioral feature vectors that appear in each risk scenario by statistically analyzing historical monitoring data, and then counts the frequency of each abnormal behavior type that accompanies the behavioral feature vectors.

[0050] For example, the formula for calculating the correlation frequency is as follows: ; In the formula: The correlation frequency between feature vector F and anomalous behavior type A; The number of times feature vector F occurs simultaneously with anomalous behavior type A; : The total number of times the feature vector F appears in the corresponding risk scenario.

[0051] Establish positive association rules between behavioral characteristics and abnormal behavior types under positive conditions; Specifically, the establishment of association rules between behavioral characteristics and abnormal behavior types under positive conditions includes: Based on the risk scenario, the behavioral characteristic sequence of the workers is segmented into segments to obtain behavioral characteristic segments; In practice, this embodiment uses a fixed time window (e.g., 10 seconds) or a fixed length window to divide the behavioral feature sequence into several independent segments for subsequent rule establishment.

[0052] Extract the abnormal behavior type corresponding to each behavioral feature segment, and establish a preliminary association between the behavioral feature segment and the abnormal behavior type; Specifically, this embodiment uses historical data labeling or expert manual labeling methods to determine whether each behavioral feature segment is accompanied by an abnormal behavior type, thereby obtaining a preliminary set of correlation relationships.

[0053] Calculate the confidence level of the preliminary association results to determine the positive association rules between high-confidence behavioral features and abnormal behavior types; In practice, this embodiment uses the following confidence calculation method to determine the positive association rules with high confidence: ; In the formula: Feature fragments Confidence of positive association with anomalous behavior type A; Feature fragments The number of times that abnormal behavior type A occurs simultaneously; Feature fragments Total number of times it appears in risky scenarios.

[0054] The behavioral feature segments whose confidence level does not meet the threshold requirement are excluded, while the behavioral feature segments that meet the threshold requirement are retained, and the positive association rule is output. In this embodiment, a confidence threshold (e.g., 0.8) is set. Behavioral feature segments with a confidence level higher than the threshold are selected as positive association rules with abnormal behavior types, while those with a confidence level lower than the threshold are excluded.

[0055] Reverse verification to determine whether the behavioral characteristics corresponding to the occurrence of abnormal behavior types are stably correlated; In specific implementation, the "reverse verification" process described in this embodiment refers to performing reverse analysis on abnormal behavior types in historical monitoring data to determine whether the corresponding behavioral characteristics consistently and stably accompany the occurrence of such abnormal behavior types, thereby evaluating whether the previously established positive association rules possess symmetry characteristics. The specific implementation method is as follows: First, a specific type of abnormal behavior (e.g., "unauthorized entry into equipment danger zones") in historical monitoring data is used as the verification object. All time points in which this type of abnormal behavior occurs are extracted from the historical monitoring data, and the behavioral feature vectors within the corresponding time segments are traced back one by one.

[0056] Secondly, the proportion of corresponding behavioral feature vectors appearing each time an abnormal behavior type occurs is statistically analyzed to determine the confidence level (strength of reverse association) of the reverse association between the abnormal behavior type and the behavioral feature vector. The calculation method for the confidence level of reverse association is as follows: ; In the formula: Abnormal behavior type A: Reverse correlation behavior feature fragment Confidence level; Abnormal behavior type A is accompanied by behavioral characteristic fragments. The number of times they occur simultaneously; : Total number of times abnormal behavior type A occurs.

[0057] Subsequently, this embodiment further determines whether the reverse association is stable based on a preset confidence threshold (e.g., 0.8). If the confidence of the reverse association exceeds the threshold, it indicates that the association between the behavioral feature and the abnormal behavior type is a stable symmetrical association; if the confidence of the reverse association is lower than the threshold, it indicates an unstable, asymmetrical association.

[0058] Remove the symmetric rules that are consistently true in the reverse verification, obtain the asymmetric association rules, and identify the types of abnormal behavior based on the asymmetric association rules; In specific implementation, this embodiment obtains asymmetric association rules in the following way: First, the high-confidence positive association rules obtained above are compared with the stable symmetric association rules determined in the reverse verification above. Then, delete the rules in the forward association rules that are the same as the stable symmetric association rules determined in the reverse verification, thereby obtaining the final set of asymmetric association rules; The set of asymmetric association rules can be represented as: In the formula, A set of asymmetric association rules; : A set of positive association rules; : The set of stable symmetry rules confirmed in reverse verification.

[0059] It should be further noted that this embodiment uses the aforementioned asymmetric association rule set for real-time identification of abnormal behavior types. Specifically, when a segment of worker behavior characteristics monitored in real time at the construction site matches a rule in the asymmetric association rule set, it is immediately identified as the corresponding abnormal behavior type, thereby achieving timely and accurate identification of abnormal behavior at the construction site.

[0060] For example, suppose an asymmetric association rule is represented as: Behavioral characteristic fragments : Feature vector of workers continuously remaining near the hazardous area of ​​equipment; Abnormal behavior type A: "Accidental entry into a hazardous area of ​​equipment"; When the construction site monitors in real time the current behavioral characteristic segment matches the aforementioned behavioral characteristic segment When a match is found, this embodiment immediately triggers the identification of abnormal behavior type A, that is, it determines that the operator is currently in an abnormal state of "accidentally entering the dangerous area of ​​the equipment", so as to initiate subsequent safety supervision measures.

[0061] S103: After identifying the abnormal behavior type, automatically retrieve the historical operating data of the construction equipment based on the risk scenario associated with the abnormal behavior type, and extract the abnormal state characteristics of the equipment that caused the abnormal behavior type from it. In a specific implementation, the extraction of device abnormal state features that trigger abnormal behavior types includes: Based on the abnormal behavior types identified in real time in step S102, determine the associated risk scenarios; It should be noted that the "risk scenario" described in this embodiment is a specific risk situation obtained based on the combination of the interaction behavior type determined in step S101 and the corresponding dangerous equipment area and personnel exposure level. The risk scenario may include, but is not limited to: personnel accidentally entering the operating area of ​​high-risk equipment, areas with drastic fluctuations in equipment load, and areas with abnormal equipment operation.

[0062] For example, when step S102 identifies the abnormal behavior type as "mistakenly entering the dangerous area of ​​the equipment", the abnormal behavior type is automatically associated with the specific "high-risk equipment operating area" risk scenario according to the risk scenario matrix defined in step S101.

[0063] Automatically retrieve historical operating data of construction equipment based on identified risk scenarios; In specific implementation, the historical operating data of construction equipment mentioned in this embodiment refers to the historical operating parameter data of construction equipment under similar or identical risk scenarios in the past. The historical operating data includes, but is not limited to: Equipment operating load data: such as boom load, hook load, hydraulic system pressure; Vibration data during equipment operation: such as vibration frequency, amplitude, and acceleration data of key components; Environmental operating data: such as ambient temperature, humidity, wind speed and direction; Equipment operation status records: such as equipment start / stop status, abnormal alarm records, and operating power fluctuation records.

[0064] Specifically, in this embodiment, data interaction is performed between the IoT data platform and the device's historical data storage system to automatically match historical data segments that are highly similar to the current risk scenario and the historical risk scenario as data sources for further analysis.

[0065] It should be further noted that automatic retrieval methods can employ time window matching or scene similarity matching: The time window matching method is as follows: Automatically retrieve historical operational data of construction equipment under the same equipment and the same work procedure based on a specific time period (e.g., 10 minutes before and after) before and after the time of abnormal behavior identification; The scene similarity matching method is as follows: Based on key indicators of risk scenarios (such as equipment load level, operating frequency range, ambient temperature and humidity, etc.), a risk scenario similarity calculation model is constructed to automatically retrieve historical operating data segments that have a similarity to the current risk scenario exceeding a threshold (e.g., 0.8).

[0066] Extract abnormal equipment status features that trigger abnormal behavior from automatically retrieved historical operating data of construction equipment; the specific implementation method is as follows: First, this embodiment uses an abnormal state feature extraction algorithm to analyze and process the automatically retrieved historical operating data of the equipment in order to identify abnormal state features in the equipment operating data.

[0067] In specific implementation, this embodiment preferably uses statistical analysis methods (such as standard deviation analysis, Z-score anomaly detection method) or signal processing methods (such as fast Fourier transform FFT, wavelet analysis) to extract abnormal state features.

[0068] For example, taking the extraction of abnormal state features based on the Z-score normalization method as an example, the specific implementation steps are as follows: First, statistical processing is performed on the automatically retrieved historical operational data sequences to calculate the mean of the data sequences. and standard deviation Calculate the Z-score: ;in, : Data anomaly score; Specific values ​​of the equipment's historical operating data; Then, define the threshold values ​​for abnormal state characteristics (e.g.) When the Z-score value in historical data exceeds a set threshold, the data is identified as an abnormal state feature.

[0069] Furthermore, taking construction equipment vibration data as an example, the Fast Fourier Transform (FFT) method is used to further analyze the equipment vibration data spectrum and extract abnormal peak values ​​of the vibration frequency spectrum as abnormal state characteristics of the equipment. The specific calculation formula for the FFT transform is as follows: In the formula, : Frequency domain characteristic spectrum of equipment vibration data; : Vibration data time-domain sampling points; N: Total number of sampling points; k: Frequency index.

[0070] After obtaining the frequency domain feature spectrum of vibration data through FFT analysis, the frequency components in the spectrum that are greater than the normal threshold are identified, and the corresponding frequency values, amplitudes, durations, etc. are recorded as abnormal state characteristics of the equipment.

[0071] S104: Based on the asymmetric association rules between abnormal equipment status characteristics and abnormal behavior types, reversely locate the potential hazards of the construction equipment, and push safety alarm information to the terminal equipment of the workers based on the potential hazards. In a specific implementation, the step of locating the potential hazards of construction equipment based on the asymmetric association rules between abnormal equipment state characteristics and abnormal behavior types includes: Establish an initial correlation between abnormal equipment status characteristics and potential equipment hazards; The establishment of the initial correlation between abnormal equipment status characteristics and potential equipment hazards includes: The key components of the equipment are classified according to their structural hierarchy. In practical implementation, this embodiment divides the construction equipment into multiple levels based on its mechanical structure, functional modules, and operating characteristics, resulting in a set of key components. For example, for a tower crane, the key components include, but are not limited to: the boom structure, the slewing mechanism, the hoisting mechanism (winch mechanism), the hydraulic or electric drive system, the base structure, and the support connection parts.

[0072] Collect operational data of different key components under normal and abnormal conditions; It should be further noted that, in this embodiment, corresponding status monitoring sensors are deployed at each of the aforementioned key components of the construction equipment, including but not limited to vibration sensors, temperature sensors, current and voltage sensors, and pressure sensors, to collect and store equipment operation data in real time. Normal condition data: Data collected by the equipment under normal daily operating conditions, used to determine the normal operating characteristics of the parts; Abnormal status data: Data collected when equipment malfunctions or is in an abnormal state, used to determine abnormal characteristics.

[0073] For example, a triaxial vibration sensor is installed on the crane boom to collect vibration acceleration data under normal and abnormal operating conditions and store it as historical operating data.

[0074] Extract the differential features between the operational data of key components and establish the initial association between the differential features and the key components; In specific implementation, this embodiment analyzes the above-mentioned normal and abnormal state data and extracts features that can clearly distinguish between normal and abnormal states as differential features. The specific implementation method is as follows: First, using statistical analysis methods, such as the rate of change method, calculate the magnitude of the difference between outlier and normal data: ; In the formula: Data differences and characteristics; : The value of the corresponding running parameter under abnormal conditions; : The value of the corresponding operating parameter under normal conditions.

[0075] Furthermore, a difference threshold (e.g., ±10%) is set, and difference data exceeding this threshold are defined as key difference features. The correlation between these difference features and the corresponding key parts is initially established.

[0076] The initial correlation was verified multiple times using historical data of construction equipment operation to form a stable initial correlation between abnormal equipment status characteristics and potential equipment locations. In practice, this embodiment involves repeated statistical verification of historical data, specifically including: Statistical analysis of the frequency of occurrence of each differential feature in different abnormal events; When the frequency of the association between the differential feature and a specific key part exceeds a preset threshold (e.g., 0.8), this association is established as a stable initial association. The above stable relationships are used to form an initial relationship database, which is then stored in the database for future use in real-time reverse location of potential hazards in the equipment.

[0077] When an abnormal behavior occurs, the initial potential equipment hazard location is traced back based on the characteristics of the abnormal equipment state; the specific implementation method is as follows: In this embodiment, during real-time monitoring, once an abnormal behavior type is identified in step S102 and the corresponding abnormal equipment status features are extracted in step S103, the initial association database constructed above is used to automatically match the current abnormal status features with the initial association of the corresponding hidden danger location, thereby tracing back and obtaining preliminary information on the hidden danger location of the equipment.

[0078] For example, when the abnormal equipment status characteristic is detected as "abnormal increase in boom vibration acceleration", the system automatically retrieves matching results from the initial association database and finds that the equipment hazard location associated with this abnormal feature is "boom structure connection". In practical applications, the same abnormal status characteristic may correspond to multiple candidate equipment hazard locations. In order to accurately locate the specific hazard location, this embodiment further combines the currently identified abnormal behavior type to further screen and confirm multiple candidate hazard locations.

[0079] The correlation between potential equipment defects and abnormal behavior types is positively verified to determine whether the defects in the equipment are stable enough to trigger abnormal behavior types. The specific implementation method is as follows: This embodiment analyzes historical data to calculate the positive correlation frequency of specific abnormal behavior types occurring when a specific equipment defect location is abnormal, in order to assess the stability of this correlation. The calculation formula is as follows: ; In the formula: : Confidence of positive correlation between the potential equipment location P and the abnormal behavior type A; The number of times abnormal behavior type A occurs when the hidden danger location P is abnormal; : The total number of times the potential hazard location P exhibits an abnormal state.

[0080] Furthermore, this embodiment sets a confidence threshold for the correlation stability (e.g., 0.8). When the calculated confidence of the positive correlation is greater than the threshold, it is considered that there is a stable positive correlation between the hidden danger location and the abnormal behavior type.

[0081] Remove the symmetric correlation that is positively verified to be stable, obtain the asymmetric correlation between the abnormal state characteristics of the equipment and the abnormal behavior type, and use the asymmetric correlation to reverse locate the hidden parts of the construction equipment. In this embodiment, the initial association traced above is compared with the stable symmetric association confirmed in the forward verification above. If the initial association is also a stable symmetric association in the forward verification, it is deleted.

[0082] Ultimately, this embodiment retains only the association relationships with asymmetric features, and uses these asymmetric association rules to reverse locate the current abnormal state features of the equipment, thereby accurately identifying the specific hidden dangers of the construction equipment.

[0083] Based on the results of reverse positioning, this embodiment automatically pushes corresponding safety alarm information to the smart terminal devices (such as smart safety helmets or smart wristbands) of workers through the Internet of Things communication network at the construction site, so as to remind workers to evacuate from the risk area in time or take corresponding safety measures, thereby ensuring the safety of personnel at the construction site.

[0084] Example 2 like Figure 2 As shown in the example, the parts not detailed in this embodiment are as shown in Example 1. This embodiment discloses an Internet of Things-based engineering project safety monitoring system, including: The scenario segmentation module 201 is used to determine the type of interaction behavior based on the real-time location information of the workers and the operating status of the construction equipment, and to segment risk scenarios based on the type of interaction behavior. The behavior recognition module 202 is used to extract the behavior features of workers based on risk scenarios using a convolutional neural network that integrates sparse perception, and to identify abnormal behavior types based on the asymmetric association rules between the behavior features and the risk scenarios. The feature extraction module 203 is used to automatically retrieve historical operating data of construction equipment based on the risk scenarios associated with the abnormal behavior type after the abnormal behavior type is identified, and extract the abnormal state features of the equipment that caused the abnormal behavior type from it. The safety management module 204 is used to reverse locate the potential hazards of the construction equipment based on the asymmetric association rules between the abnormal status characteristics and abnormal behavior types of the equipment, and to push safety alarm information to the terminal equipment of the operators based on the potential hazards.

[0085] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for safety supervision of engineering projects based on the Internet of Things, characterized in that, include: The types of interactive behaviors are determined based on the real-time location information of the workers and the operating status of the construction equipment, and risk scenarios are classified according to the types of interactive behaviors. Based on risk scenarios, we use convolutional neural networks that integrate sparse perception to extract the behavioral features of workers, and identify abnormal behavior types based on the asymmetric association rules between behavioral features and risk scenarios. Once the abnormal behavior type is identified, the historical operating data of the construction equipment is automatically retrieved based on the risk scenario associated with the abnormal behavior type, and the abnormal state characteristics of the equipment that caused the abnormal behavior type are extracted from it. Based on the asymmetric association rules between abnormal equipment status characteristics and abnormal behavior types, the potential hazards of the construction equipment are located in reverse, and safety alarm information is pushed to the terminal equipment of the operators according to the potential hazards.

2. The method for safety supervision of engineering projects based on the Internet of Things according to claim 1, characterized in that, The method of determining the interaction behavior type based on the real-time location information of the workers and the operating status of the construction equipment, and classifying risk scenarios based on the interaction behavior type, includes: Obtain the real-time relative position of the workers to the construction equipment; Based on the trend of the duration of the construction equipment's operation, determine the current operating stage of the construction equipment; Based on the combination of the relative positional relationship of the workers and the operational phase of the construction equipment, the types of interactive behaviors between the workers and the construction equipment are determined. Based on the combination of the dangerous areas of the equipment and the degree of personnel exposure involved in the interaction behavior type, the corresponding risk scenarios are divided.

3. The method for safety supervision of engineering projects based on the Internet of Things according to claim 2, characterized in that, The sparse-aware convolutional neural network includes: The feature sparsification convolution module uses dilated convolution to extract behavioral feature maps with sparse properties from consecutive frame images. The feature difference enhancement module highlights the local difference characteristics of abnormal behavior regions in the behavior feature map through local differential encoding; The attention weighting module generates spatial location weights based on the local differences in regions of abnormal behavior. The feature fusion module performs element-wise weighted fusion of different regions of the behavior feature map based on spatial location weights to obtain behavioral features that can represent the abnormal behavior types of workers.

4. The method for safety supervision of engineering projects based on the Internet of Things according to claim 3, characterized in that, The structure of the sparse sensing module includes: The sparse weight calculation unit is used to calculate the sparse weights of each feature channel based on the response intensity distribution of each channel in the convolutional feature. The feature channel selection unit is connected to the sparse weight calculation unit and is used to select high-weight feature channels according to the sparse weight. A sparse mask generation unit, connected to the feature channel selection unit, is used to generate a feature mask based on the selected feature channel. The sparse feature output unit, connected to the sparse mask generation unit, is used to apply a feature mask to the original convolutional features and output a sparse feature map.

5. The method for safety supervision of engineering projects based on the Internet of Things according to claim 4, characterized in that, The method for obtaining the sparse-aware convolutional neural network includes: Construct training samples of worker behavior characteristics and extract preliminary convolutional features using an initial convolutional neural network model; Based on the sparse distribution of channel responses in the initial convolutional features, a sparse target mask is generated for constrained training. A sparse target mask is introduced into the model training loss function, and the network weights are updated based on the backpropagation algorithm. Repeat the training iterations until the sparsity of the behavioral features output by the network reaches the requirements of the sparse target mask, and obtain a convolutional neural network that integrates sparse perception.

6. The method for safety supervision of engineering projects based on the Internet of Things according to claim 5, characterized in that, The method for identifying abnormal behavior types based on asymmetric association rules between behavioral characteristics and risk scenarios includes: Obtain the correlation frequency between worker behavior characteristics and abnormal behavior types in different risk scenarios; Establish association rules between behavioral characteristics and abnormal behavior types under positive conditions; Reverse verification to determine whether the behavioral characteristics corresponding to the occurrence of abnormal behavior types are stably correlated; Remove the symmetric rules that are consistently true in the reverse verification, obtain the asymmetric association rules, and identify the types of abnormal behavior based on the asymmetric association rules.

7. The method for safety supervision of engineering projects based on the Internet of Things according to claim 6, characterized in that, The establishment of association rules between behavioral characteristics and abnormal behavior types under positive conditions includes: Based on the risk scenario, the behavioral characteristic sequence of the workers is segmented into segments to obtain behavioral characteristic segments; Extract the abnormal behavior type corresponding to each behavioral feature segment, and establish a preliminary association between the behavioral feature segment and the abnormal behavior type; Calculate the confidence level of the preliminary association results to determine the positive association rules between high-confidence behavioral features and abnormal behavior types; The behavioral feature segments whose confidence level does not meet the threshold requirement are excluded, while the behavioral feature segments that meet the threshold requirement are retained, and the positive association rule is output.

8. The method for safety supervision of engineering projects based on the Internet of Things according to claim 7, characterized in that, The method of locating potential hazards in construction equipment based on asymmetric association rules between abnormal equipment state characteristics and abnormal behavior types includes: Establish an initial correlation between abnormal equipment status characteristics and potential equipment hazards; When an abnormal behavior occurs, trace the initial potential equipment location based on the characteristics of the abnormal equipment status; Positively verify the correlation between the location of potential equipment hazards and the type of abnormal behavior to determine whether the location of potential equipment hazards is stable and triggers the type of abnormal behavior. Remove the symmetric correlation that is positively verified to be stable, obtain the asymmetric correlation between the abnormal state characteristics of the equipment and the abnormal behavior type, and use the asymmetric correlation to reverse locate the hidden parts of the construction equipment.

9. The method for safety supervision of engineering projects based on the Internet of Things according to claim 8, characterized in that, The establishment of the initial correlation between abnormal equipment status characteristics and potential equipment hazards includes: The key components of the equipment are classified according to their structural hierarchy. Collect operational data of different key components under normal and abnormal conditions; Extract the differential features between the operational data of key components and establish the initial association between the differential features and the key components; By using historical data from the operation of construction equipment, the initial correlation was verified multiple times to establish a stable initial correlation between abnormal equipment status characteristics and potential equipment locations.

10. An Internet of Things (IoT)-based engineering project safety monitoring system, implemented based on any one of claims 1-9, characterized in that, include: The scenario segmentation module is used to determine the type of interaction behavior based on the real-time location information of the workers and the operating status of the construction equipment, and to classify risk scenarios according to the type of interaction behavior. The behavior recognition module is used to extract the behavioral features of workers based on risk scenarios using a convolutional neural network that integrates sparse perception, and to identify abnormal behavior types based on the asymmetric association rules between behavioral features and risk scenarios. The feature extraction module is used to automatically retrieve historical operating data of construction equipment based on the risk scenarios associated with the abnormal behavior type after the abnormal behavior type is identified, and extract the abnormal state features of the equipment that caused the abnormal behavior type from it. The safety management module is used to reverse locate the potential hazards of construction equipment based on the asymmetric association rules between the characteristics of abnormal equipment status and the types of abnormal behavior, and to push safety alarm information to the terminal equipment of the operators based on the potential hazards.

Images