A power construction behavior safety detection method and system

Abstract

This application discloses a method and system for safety detection of power construction activities, relating to power safety: collecting historical power work order data to construct a work order risk feature database; training a risk identification model based on multi-task learning according to the work order risk feature database; collecting text data of the current power work order, extracting the work content and safety measure fields, and generating text feature vectors; using a discretized sea squirt optimization algorithm to search for the optimal parameter configuration in the parameter space of the risk identification model; collecting real-time image data of the work site, performing target detection and feature extraction on the image data, identifying the coordinates of the workers' positions, as well as the type and wearing status of safety protective equipment, encoding the detection results into structured data, and generating image feature vectors. Addressing the lack of a dynamic correlation verification mechanism between the existing power work order text specifications and actual work site behavior, this application transforms the static requirements of the work order into dynamic behavioral supervision standards.

Description

Technical Field

[0001] This application relates to the field of power safety, and in particular to a method and system for safety detection of power construction activities. Background Technology

[0002] With the continuous expansion of large-scale engineering projects and the increasing mechanization of construction, the safety management of aerial work platforms has become a crucial link in ensuring project quality and the safety of workers. Statistics show that over 70% of aerial work accidents are related to violations of operating procedures, inadequate protective measures, and insufficient control over platform stability. Traditional aerial work safety management mainly relies on work instructions and manual inspections, using pre-established written regulations to constrain work behavior. However, this static management model reveals many shortcomings in the dynamic and complex environment of aerial work.

[0003] Work instructions, as the core system for safety management of high-altitude operations, detail the requirements for work content, safety measures, operating procedures, and hazard pre-control. However, there is a significant temporal and spatial separation between the textual specifications in the work instructions and the actual execution on-site. On the one hand, work instructions are formulated before construction, making it difficult to foresee all risk scenarios during 24-meter-high operations using boom lifts or multi-person collaborative operations using scissor lifts. On the other hand, subjective differences exist in the understanding and execution of the specifications by on-site workers, leading to frequent instances of "disregard for regulations and distorted implementation." Although some engineering companies have introduced video monitoring systems, most remain at the post-event review stage, lacking real-time behavior recognition and risk warning capabilities.

[0004] In recent years, with the development of artificial intelligence technology, deep learning-based image recognition technology has begun to be applied to the field of high-altitude work safety monitoring. Existing technologies use target detection algorithms to identify elements such as workers, safety equipment, high-altitude work platforms, and surrounding obstacles, achieving a certain degree of automated monitoring of the work site. However, these methods generally suffer from the following technical shortcomings:

[0005] First, there is a lack of an effective mechanism for linking textual specifications with image recognition. The existing system treats work instruction management and on-site monitoring as two separate modules, failing to establish a dynamic mapping relationship from textual requirements to behavioral standards. This results in image recognition results that cannot accurately reflect whether the safety specifications for specific types of aerial work (such as linear lifting operations on boom lifts and large-scale collaborative operations on scissor lifts) are met.

[0006] Secondly, the parameter configuration of risk identification models lacks targeted optimization. Different types of high-altitude operations (such as using a boom lift equipped with an active protection system for 460kg heavy-duty operations, using a scissor lift for multi-person collaborative operations, and using an extension platform on an inclined wall) have different risk characteristics and monitoring priorities, but existing methods mostly use uniform model parameters, which cannot adapt to the diverse needs of different operational scenarios. Even some studies that attempt to introduce parameter optimization face the problem of mismatch between optimization algorithms and parameter spaces.

[0007] Especially in parameter optimization, traditional continuous optimization algorithms (such as particle swarm optimization and genetic algorithms) generate a large number of invalid intermediate solutions when dealing with discrete parameters (such as batch size and number of network layers) in deep learning models. For example, when an optimization algorithm generates a continuous value with a batch size of 31.7, it needs to be converted to 32 by simple rounding. This crude discretization method destroys the search logic of the optimization algorithm, causing it to get stuck in local optima or converge slowly. While the tunic optimization algorithm, as a novel swarm intelligence algorithm, performs well in continuous optimization problems, its position update mechanism is also based on a continuous space design. When directly applied to discrete parameter optimization, it suffers from low search efficiency and unstable solution quality.

[0008] Furthermore, the application of existing multi-task learning frameworks in the safety inspection of aerial work operations is not yet in-depth. Aerial work safety involves multiple risk dimensions, such as platform operation procedures, working height control, wearing of protective equipment, and platform stability monitoring. These dimensions are both independent and interconnected. Simple single-task models cannot capture this complex risk structure, while existing multi-task learning methods lack task design and loss function optimization tailored to the characteristics of aerial work platforms (such as reach control for boom lifts and load distribution for scissor lifts).

[0009] In summary, existing technologies struggle to effectively translate the textual specifications of work instructions into on-site monitoring of high-altitude operations. They also fail to dynamically adjust monitoring strategies based on different work platform types and work heights. Furthermore, the parameter optimization process suffers from technical bottlenecks due to the mismatch between algorithm mechanisms and parameter spaces, severely hindering the improvement of intelligent safety management in large-scale machinery projects. Summary of the Invention

[0010] In view of the lack of a dynamic correlation verification mechanism between the text specifications of power work permits and the actual behavior at the work site in the existing technology, this application provides a method and system for safety detection of power construction behavior, which constructs a dual verification mechanism of text analysis and image verification, and transforms the static requirements of work permits into dynamic behavioral supervision standards.

[0011] One aspect of this application provides a method for safety detection of power construction activities, comprising: S1, collecting historical power work order data, and extracting fields such as work content, safety measures, work steps, and hazard pre-control measures from the work orders; constructing a work order risk feature library based on the extracted fields, wherein each record in the risk feature library contains a work type code, a risk feature vector, and a historical accident marker; S2, training a risk recognition model based on multi-task learning based on the work order risk feature library, wherein the risk recognition model includes a shared feature extraction layer and multiple task output heads, each output head corresponding to a type of work risk assessment task; S3, collecting text data of the current power work order. The process involves: S1) Extracting the work content and safety measures fields to generate text feature vectors; S2) Using the discretized sea squirt optimization algorithm to search for the optimal parameter configuration in the parameter space of the risk identification model to obtain the optimal model parameters that match the current work order type; S3) Collecting real-time image data of the work site, performing target detection and feature extraction on the image data, identifying the coordinates of the workers' positions, as well as the type and wearing status of safety protective equipment, encoding the detection results into structured data, and generating image feature vectors; S4) Using the image feature vectors and the optimal model parameters, using the risk identification model to make predictions and obtain risk assessment results.

[0012] Furthermore, in S2, based on the work permit risk feature library, a risk identification model based on multi-task learning is trained. This model includes a shared feature extraction layer and multiple task output heads, each corresponding to a type of work risk assessment task. These tasks include: extracting risk feature vectors and historical accident markers from the work permit risk feature library; dividing the dataset into three subsets based on work type encoding: live-line work, switching operations, and maintenance work; constructing a multi-task learning network architecture based on these subsets; where the shared feature extraction layer uses a three-layer fully connected network; and setting three task output heads: an operation sequence compliance output head to predict whether switching operation steps conform to standard operating procedures; a safety distance detection output head to predict whether the safety distance between the operator and live equipment meets regulatory requirements; and a protective equipment compliance output head to predict whether the use of safety helmets and insulating gloves meets regulatory requirements. A multi-task loss function is set, and the batch gradient descent algorithm is used for model training to obtain the risk identification model based on multi-task learning.

[0013] Furthermore, the expression for the multi-task loss function is as follows: ;in, The binary cross-entropy loss is for compliance with the order of operations. For multi-class cross-entropy loss in safe distance detection, To protect against the standard binary cross-entropy loss of equipment, This refers to the task weighting coefficient;

[0014] Further, in step S3, the text data of the current power work order is collected, and the fields of work content and safety measures are extracted to generate a text feature vector. Based on the text feature vector, the optimal parameter configuration is searched in the parameter space of the risk identification model using the discretized tunicate optimization algorithm to obtain the optimal model parameters that match the current work order type. These parameters include: the initial parameters of the discretized tunicate optimization algorithm, which include: the tunicate population size N, and the maximum number of iterations. and temperature parameters Define the parameter search space, including: learning rate Batch size and dropout rate Construct a discrete index mapping table to map the continuous parameter search space to discrete indices; construct the fitness function F: ;in, The accuracy of the model on the validation set. For model complexity, To balance the coefficients; perform a discretized tunicate optimization process, iteratively optimizing until the termination condition is met, and output the optimal parameter configuration. .

[0015] In particular, the position update mechanism of the traditional Squirting Optimization Algorithm (SSA) is based on continuous mathematical operations, and its leader position update formula... ) and follower position update formula Essentially, it involves interpolation and extrapolation operations in a continuous space. However, the key hyperparameters of risk identification models exhibit significant discrete characteristics: batch size must be a positive integer (e.g., 16, 32, 64), learning rate is typically selected from a predefined set (e.g., 0.001, 0.01, 0.1), and dropout rate often uses fixed intervals (e.g., 0.1, 0.2, 0.3). Simply rounding the continuous optimization results directly will disrupt the algorithm's search logic, leading to suboptimal solutions.

[0016] This application introduces a probabilistic discrete mapping mechanism, using temperature parameters... The controlled probability distribution enables an intelligent transformation from a continuous search space to a discrete decision space. This design, while maintaining the advantages of SSA (Swarm Intelligence Search), solves the problem of discrete parameter optimization.

[0017] Furthermore, S35, execute the discretized tunicate optimization process, including: initializing the population data matrix. Where N is the population size, and each row represents an individual in a three-dimensional continuous space. Location data in the middle;

[0018] Update the position data in each iteration: for the leader individual, according to the formula... Update continuous position data; for individual followers, according to the formula Calculate new location data; for each consecutive value Calculate candidate discrete indices: ; Calculate the mapping distance: According to probability Perform random mapping to generate the final discrete index; extract actual parameter values ​​from the parameter candidate set based on the discrete index; configure the model using the extracted parameter values, evaluate performance on the validation set, calculate fitness values, and update the optimal solution record;

[0019] In particular, the optimization of hyperparameters in deep learning models faces a fundamental contradiction: advanced swarm intelligence optimization algorithms (such as the tunic algorithm) rely on mathematical operations in continuous space to achieve efficient search, while key parameters in practical applications (learning rate, batch size, dropout rate) must take discrete values. Traditional post-processing rounding methods not only disrupt the search logic of the optimization algorithm, but may also cause parameter configurations to deviate from the optimal region, creating safety hazards in high-risk power operation scenarios.

[0020] This application, based on probabilistic discrete mapping for temperature control, achieves intelligent bridging of the continuous search space and the discrete parameter space through a three-stage mechanism of continuous exploration – probabilistic mapping – discrete decision-making. That is, this application does not simply quantize continuous values ​​into the nearest discrete point, but rather uses a probability function... This gives the algorithm the ability to make intelligent selections from a discrete set of candidates.

[0021] Specifically, when the algorithm finds position c in the continuous space [0, 1], the system first calculates its distance to the nearest discrete index. Then based on temperature parameters The probability distribution of the control determines the final discrete mapping result. The closer the distance, the higher the mapping probability, but it is not absolutely certain. The introduction of the temperature parameter gives the algorithm different exploration characteristics at different optimization stages.

[0022] The core advantage of the tunic algorithm lies in its chain-like collaborative search mechanism: the leader, through... Explore new areas, followers through This enables information transfer and local refinement. If these operations are performed directly in discrete space, position averaging will lose its meaning, the chain structure will break, and the algorithm will degenerate into a simple random search.

[0023] The introduction of the probability mapping mechanism in this application ensures that all the desirable properties of the algorithm are preserved in continuous space: the population remains in... The search space evolves according to the original formula, the search trajectory remains continuous, and the information transmission mechanism functions normally. Discrete parameter values ​​are only obtained through probability mapping when fitness needs to be evaluated.

[0024] Furthermore, power system risks exhibit nonlinear and abrupt characteristics. Fine-tuning the learning rate from 0.01 to 0.012 may have no effect, but misadjusting it to 0.1 could lead to gradient explosion in the model, completely destroying its risk identification ability. Probabilistic mapping ensures that parameters can only take predefined safe values, fundamentally avoiding the risk of parameter avalanche.

[0025] Furthermore, temperature parameters The annealing mechanism aligns perfectly with the risk-based management concept of power operations. When handling low-risk operations such as routine maintenance, a higher temperature can be set, allowing the algorithm to explore boldly and find more efficient parameter combinations. However, when handling high-voltage live-line work, lowering the temperature allows the algorithm to quickly converge to the most stable parameter configuration, reflecting the industry principle that the higher the safety level, the smaller the fault tolerance margin.

[0026] Furthermore, in step S4, real-time image data of the work site is collected, target detection and feature extraction are performed on the image data, the coordinates of the workers' positions, and the types and wearing status of safety protective equipment are identified. The detection results are encoded into structured data to generate image feature vectors. This includes: collecting real-time image data of the work site and preprocessing the images; using the YOLOv5 target detection model to identify key targets in the images: detecting workers and outputting bounding box coordinates. and confidence level The system detects safety helmets and insulating gloves, outputting the bounding boxes and confidence levels of the corresponding protective equipment; it detects live equipment and working equipment, outputting the bounding box coordinates and category labels of the equipment; it performs quantitative analysis of the spatial relationships and wearing status of the targets based on the bounding box coordinates to obtain the detection results; and it encodes the detection results into structured data to generate image feature vectors.

[0027] Furthermore, the spatial relationship and wearing status of the target are quantitatively analyzed based on the bounding box coordinates to obtain the detection results, including: extracting edge sampling point sets from the bounding boxes of the workers and the live equipment respectively; calculating the Euclidean distance matrix between point pairs based on the edge sampling point sets; extracting the minimum value from the Euclidean distance matrix as the safety distance, which is used as the target spatial relationship detection result; dividing the bounding box of the workers into functional areas according to a preset ratio; calculating the overlap between the bounding box of the protective equipment and the corresponding human functional area; and generating a binarized wearing status detection result based on the overlap threshold.

[0028] In particular, traditional object detection algorithms use rectangular bounding boxes as output, a design stemming from considerations of computational efficiency and ease of annotation. However, this abstraction presents several problems in power operation scenarios. First, there is an irreconcilable contradiction between the dynamic complexity of the human body contour and the static rectangular bounding box. Workers' postures vary greatly during tasks—limbs outstretched when climbing, arms extended forward when operating, and bending over during maintenance. These posture changes cause the actual space occupied by the human body to far exceed or deviate from the bounding box, inevitably leading to significant errors in distance calculations based on the box's center or corners. Second, the irregular shape of power equipment exacerbates this problem. The cylindrical body of a transformer, the protruding operating handle of a switchgear, and the slender structure of busbars—these complex shapes, when forcibly compressed into rectangular bounding boxes, lose a significant amount of crucial spatial information.

[0029] Therefore, this application proposes an edge sampling point set + Euclidean distance matrix to expand the target representation from four corner points to dozens of edge points. Specifically, uniform sampling is performed on the four edges of each bounding box, with a typical sampling density of 10-20 points per edge. Euclidean distance matrix The algorithm extends distance calculation from one-to-one to many-to-many. For m sampling points of workers and n sampling points of live equipment, the algorithm calculates the distance between all possible point pairs: Then extract the minimum value from the matrix. As a safety distance, this global search strategy ensures that the system can capture the most dangerous distance information regardless of the object's deformation or the angle from which it approaches. Furthermore, the dynamic nature of power operations requires the monitoring system to respond to attitude changes in real time. Edge sampling points automatically adjust their positions as the bounding box updates, ensuring that the target's critical contours are always covered.

[0030] Furthermore, in step S5, based on the image feature vector and the optimal model parameters, a risk identification model is used to make predictions and obtain risk assessment results. This includes: configuring the risk identification model with the optimal model parameters to establish a parameterized model instance; using the image feature vector as input data, and according to the parameterized model instance: performing a nonlinear transformation on the input features through a shared feature extraction layer; inputting the transformed features in parallel into multiple task output heads; extracting the corresponding prediction probability distribution from each output head; performing threshold judgment and category mapping based on the prediction probability distribution; and generating risk assessment results through weighted aggregation based on the threshold judgment and category mapping results.

[0031] Another aspect of this application provides a safety monitoring system for power construction activities, comprising: a historical data processing module for collecting historical power work order data and extracting fields such as work content, safety measures, work steps, and hazard pre-control measures from the work orders; constructing a work order risk feature library based on the extracted fields, wherein each record in the risk feature library contains a work type code, a risk feature vector, and a historical accident marker; a model training module connected to the historical data processing module for training a risk recognition model based on multi-task learning according to the work order risk feature library, wherein the risk recognition model includes a shared feature extraction layer and multiple task output heads, each output head corresponding to a type of work risk assessment task; and a parameter optimization module for collecting text data from the current power work order and optimizing the data. The system extracts the work content and safety measures fields to generate text feature vectors. Based on these text feature vectors, it uses a discretized sea squirt optimization algorithm to search for the optimal parameter configuration in the parameter space of the risk identification model, obtaining the optimal model parameters that match the current work order type. An image processing module collects real-time image data from the work site, performs target detection and feature extraction on the image data, identifies the coordinates of the workers' positions, and identifies the type and wearing status of safety protective equipment. The detection results are encoded into structured data to generate image feature vectors. A risk assessment module, connected to the model training module, parameter optimization module, and image processing module, predicts risks based on the image feature vectors and the optimal model parameters using the risk identification model, obtaining the risk assessment results.

[0032] Compared to existing technologies, the advantages of this application are:

[0033] To address the shortcomings of existing technologies, such as the lack of a dynamic correlation verification mechanism between the textual specifications of power work permits and actual on-site behaviors, and the problem that traditional continuous optimization algorithms generate a large number of invalid candidate solutions when processing the discrete parameter space of risk identification models, thus failing to quickly adapt to the real-time safety monitoring needs of different work types, this application provides a method and system for safety detection of power construction behaviors. This method automatically converts the static safety requirements in the work permit into dynamic monitoring parameters that match the current work scenario using a discretized optimization algorithm. It establishes a closed-loop detection mechanism of textual specifications – parameter optimization – image verification. While ensuring parameter search efficiency, it achieves accurate identification and real-time early warning of multi-dimensional risk factors such as operation sequence, safety distance, and protective equipment, effectively solving the problem of the disconnect between regulatory requirements and behavioral execution in power construction safety supervision. Attached Figure Description

[0034] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0035] Figure 1 This is an exemplary flowchart of a safety detection method for power construction activities according to some embodiments of this application;

[0036] Figure 2 This is an exemplary flowchart illustrating the construction of a multi-task learning risk identification model according to some embodiments of this application;

[0037] Figure 3 This is an exemplary flowchart illustrating the acquisition of optimal parameter configuration according to some embodiments of this application;

[0038] Figure 4 This is an exemplary flowchart illustrating the construction of image feature vectors according to some embodiments of this application. Detailed Implementation

[0039] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0040] like Figure 1 As shown, S1 involves collecting historical power work order data and extracting fields for work content, safety measures, work steps, and hazard pre-control measures from the work orders; constructing a work order risk feature library based on the extracted fields, where each record contains a work type code, a risk feature vector, and historical accident markers; S2 involves training a multi-task learning-based risk identification model based on the work order risk feature library, where the risk identification model includes a shared feature extraction layer and multiple task output heads, each corresponding to a type of work risk assessment task; S3 involves collecting the text data of the current power work order and extracting the work content and safety measures fields. S1. Generate text feature vectors; based on the text feature vectors, use the discretized sea squirt optimization algorithm to search for the optimal parameter configuration in the parameter space of the risk identification model to obtain the optimal model parameters that match the current work order type; S4. Collect real-time image data of the work site, perform target detection and feature extraction on the image data, identify the location coordinates of the workers, as well as the type and wearing status of safety protective equipment, encode the detection results into structured data, and generate image feature vectors; S5. Based on the image feature vectors and the optimal model parameters, use the risk identification model to make predictions and obtain the risk assessment results.

[0041] Specifically, S1 involves collecting historical power work order data and extracting fields related to work content, safety measures, work steps, and hazard prevention and control measures from the work orders; based on the extracted fields, a work order risk feature database is constructed.

[0042] S11. Perform text preprocessing on the extracted fields of work content, safety measures, work steps and hazard point pre-control measures, including removing special characters, word segmentation and stop word filtering, to obtain preprocessed text data;

[0043] S12. Use the TF-IDF algorithm to extract features from the preprocessed text data, calculate the weight values ​​of each word in different fields, and construct an initial feature matrix.

[0044] S13. Perform principal component analysis to reduce the dimensionality of the initial feature matrix, retain the principal components whose cumulative variance contribution rate reaches the preset threshold, and generate the dimensionality-reduced risk feature vector.

[0045] S14. Based on the keywords in the job content field, match the predefined job type dictionary to determine the job type code corresponding to each work order;

[0046] S15. Query the historical accident database, associate the corresponding accident records with the work order number, and set the historical accident mark to the accident type code when there is an associated accident, and set the historical accident mark to the safety mark when there is no associated accident.

[0047] S16. Combine the work type code, risk feature vector, and historical accident marker into a structured record and store it in the work ticket risk feature database.

[0048] like Figure 2 As shown in Figure S2, a risk identification model based on multi-task learning is trained according to the work order risk feature library. The risk identification model includes a shared feature extraction layer and multiple task output heads, each corresponding to a type of work risk assessment task, including:

[0049] Risk feature vectors and historical accident markers are extracted from the work order risk feature database. The dataset is divided into three subsets based on the work type code: live-line work, switching operation, and maintenance work.

[0050] Based on the data subset, a multi-task learning network architecture is constructed, in which the shared feature extraction layer adopts a three-layer fully connected network with three task output heads:

[0051] The operation sequence compliance output header is used to predict whether the switching operation steps comply with the standard operating procedures.

[0052] The safety distance detection output head is used to predict whether the safety distance between the operator and the live equipment meets the requirements of the regulations.

[0053] The protective equipment specification output head is used to predict whether the use of safety helmets and insulating gloves meets the specification requirements;

[0054] By setting a multi-task loss function and using the batch gradient descent algorithm for model training, a risk identification model based on multi-task learning is obtained.

[0055] The multi-task loss function is set, including:

[0056] S31. Define the multi-task loss function expression as follows:

[0057] ;in, The binary cross-entropy loss is for compliance with the order of operations. For multi-class cross-entropy loss in safe distance detection, To protect against the standard binary cross-entropy loss of the equipment, This refers to the task weighting coefficient;

[0058] S32. For the three binary classification tasks, a unified binary cross-entropy loss calculation method is adopted:

[0059] Where N is the number of task samples, For real labels, The logit value output by the model. It is the sigmoid activation function;

[0060] S34. Set task weight coefficients according to the importance level of power construction safety: Among them, safe distance detection has the highest weight because it is directly related to personal safety.

[0061] S35. Add an L2 regularization term to prevent overfitting: Where β=0.0001 is the regularization coefficient, and W is the set of model parameters.

[0062] like Figure 3 As shown, S3 collects the text data of the current power work order, extracts the work content and safety measures fields, and generates a text feature vector;

[0063] Based on the text feature vector, the optimal parameter configuration is searched in the parameter space of the risk identification model using the discretized tunicate optimization algorithm to obtain the optimal model parameters that match the current work order type, including:

[0064] S31. Collect the text data of the current power work order, use regular expressions to extract the work content and safety measures fields, and perform word segmentation and stop word removal on the extracted text;

[0065] S32. Use the jieba word segmentation tool to segment Chinese words, remove stop words, and retain power industry terminology; use the Word2Vec model to convert the segmentation results into 300-dimensional word vectors, perform average pooling on the word vector matrix of each field to generate field-level feature vectors; concatenate the field-level feature vectors of the work content and safety measures to form a 256-dimensional text feature vector.

[0066] S33. Initialize the parameters of the discretized tunicate optimization algorithm:

[0067] Set the population size of the salps N=30 and the maximum number of iterations. Temperature parameters ;

[0068] Define the parameter search space, including the learning rate. Batch size dropout rate ;

[0069] Initialize the data structure for the discretized tunicate optimization algorithm: Create a parameter candidate set: Learning rate set The set of batch sizes BS = {16, 24, 32, 48, 64} contains 10 discrete values, and the set of dropout rates DR = {0.1, 0.2, 0.3, 0.4, 0.5} contains 5 discrete values. A discrete index mapping table is constructed to map the continuous space [0, 1] to the discrete index {0, 1, ..., D-1}.

[0070] S34. Construct the fitness function: Where accuracy is the model's accuracy on the validation set, and complexity is the model's complexity. This is the balance coefficient;

[0071] S35. Perform the discretization optimization process for *Succubus*:

[0072] Initialize population data matrix Where N is the population size, and each row represents an individual in a three-dimensional continuous space. Location data in the middle;

[0073] Update the position data in each iteration:

[0074] For individual leaders, according to the formula Update continuous location data;

[0075] For individual followers, according to the formula Calculate new location data;

[0076] A probabilistic boundary softening mechanism is used to map continuous positions to discrete parameters: for each continuous value Calculate candidate discrete indices: ; Calculate the mapping distance: According to probability Perform random mapping to generate the final discrete index;

[0077] Extract actual parameter values ​​from the candidate parameter set based on discrete indices: Learning rate: ,in, Represents the candidate set of learning rates. For the corresponding discrete index; batch size: ,in, Represents the candidate set of learning rates. For the corresponding discrete index; dropout rate: ,in, Represents the candidate set of learning rates. Update the global optimal solution for the corresponding discrete index;

[0078] S36. Iterate and optimize until the termination condition is met, and output the optimal parameter configuration. This parameter configuration is then applied to the risk identification model.

[0079] like Figure 4 As shown, S4 collects real-time image data from the work site, performs target detection and feature extraction on the image data, identifies the coordinates of the workers' positions, and identifies the types and wearing status of safety protective equipment. The detection results are encoded into structured data to generate image feature vectors.

[0080] S41. Collect real-time image data from the work site and preprocess the images:

[0081] The original images were uniformly scaled to a resolution of 640×480.

[0082] Perform histogram equalization to enhance image contrast;

[0083] A Gaussian filter is applied to remove image noise, with a filter kernel size of 3×3;

[0084] S42. Use the YOLOv5 object detection model to identify key targets in an image:

[0085] Inspection personnel, output bounding box coordinates and confidence level ;

[0086] It can detect protective equipment such as safety helmets, insulating gloves, and insulating shoes, and output the bounding boxes and confidence scores of various protective equipment.

[0087] Detects live equipment and operating equipment, and outputs the bounding box coordinates and category labels of the equipment;

[0088] S43. Calculate the spatial relationship data between the operators and the live equipment:

[0089] Extract the center point coordinates of the personnel bounding box ;

[0090] Edge sampling is performed on the bounding box of the workers, and M=20 points are uniformly sampled on the four sides to obtain the edge point set of the workers. , Represents the coordinates of the i-th sampling point;

[0091] Edge sampling is performed on the bounding box of the live equipment. N=20 points are uniformly sampled on the four edges to obtain the set of edge points of the equipment. , Represents the coordinates of the j-th sampling point;

[0092] Calculate the minimum edge distance between personnel and live equipment: ;

[0093] Convert pixel distance to actual physical distance based on image depth information or calibration parameters. ;

[0094] S44. Analyze data on the wearing status of protective equipment:

[0095] Define the area around the head of the person as the upper 1 / 3 of the bounding box;

[0096] Calculate the intersection area of ​​the helmet's bounding box and the head region. Area of ​​union ;

[0097] Calculate the intersection-union ratio: ;

[0098] when When worn, the status is marked as 1; otherwise, it is marked as 0.

[0099] S45. Encode the detection results into a structured data matrix:

[0100] Create a 12-dimensional feature vector: [Number of personnel, mean x-coordinate of personnel, mean y-coordinate of personnel, minimum safe distance, helmet wearing rate, insulated glove wearing rate, insulated shoe wearing rate, number of live equipment, work area, personnel density, equipment ratio, mean detection confidence];

[0101] The coordinate values ​​are normalized and mapped to the interval [0, 1].

[0102] The wear rate data is encoded using 0-1 encoding;

[0103] S46. Feature enhancement for structured data:

[0104] The first 8 principal components were extracted using principal component analysis (PCA);

[0105] The principal components are concatenated with the original 12-dimensional features to form a 20-dimensional image feature vector.

[0106] Perform L2 normalization on the feature vectors to ensure that the vector magnitude is 1.

[0107] In S5, based on the image feature vector and the optimal model parameters, a risk identification model is used to make predictions and obtain risk assessment results, including:

[0108] S51. Loaded Optimal Model Parameters Configure the network parameters of the risk identification model;

[0109] S52. Input the image feature vector into the configured risk identification model:

[0110] Image feature vectors undergo feature transformation through a shared feature extraction layer;

[0111] The transformed features are input into the three task output heads respectively;

[0112] S53. Obtain the prediction results from the output headers of each task:

[0113] Operation sequence compliance output header outputs binary classification probability value ;

[0114] The safe distance detection output head outputs a multi-class probability distribution. Where k is the distance level number;

[0115] Protective equipment standard output head outputs binary classification probability value ;

[0116] S54. Post-processing and risk level determination of the prediction results:

[0117] when At that time, there is a risk of violating the order of marking operations;

[0118] according to The category corresponding to the highest probability is used to determine the risk level of the safe distance;

[0119] when At that time, it was noted that protective equipment was not worn correctly;

[0120] S55. Generate a risk assessment report by integrating the prediction results of each task:

[0121] Calculate the overall risk score: ,in, Based on the risk level of safe distance, ;

[0122] Risk levels are determined based on the overall risk score: Low risk Medium risk High risk ;

[0123] The output includes a structured assessment result containing various risk indicators and a comprehensive risk level.

[0124] The foregoing illustrative description of the present application and its embodiments is not restrictive and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The accompanying drawings are only one embodiment of the present application, and the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present application, such designs should fall within the scope of protection of this application. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A safety detection method for power construction activities, characterized in that, include: S1. Collect historical power work order data and extract the fields of work content, safety measures, work steps, and hazard pre-control measures from the work order; construct a work order risk feature library based on the extracted fields. Each record in the risk feature library contains a work type code, a risk feature vector, and historical accident markers. S2, based on the work ticket risk feature library, train a risk identification model based on multi-task learning. The risk identification model includes a shared feature extraction layer and multiple task output heads, each of which corresponds to a type of work risk assessment task. S3: Collect the text data of the current power work order, extract the work content and safety measures fields, and generate a text feature vector; Based on the text feature vector, the optimal parameter configuration is searched in the parameter space of the risk identification model using the discretized tunicate optimization algorithm to obtain the optimal model parameters that match the current work order operation type. S4 collects real-time image data of the work site, performs target detection and feature extraction on the image data, identifies the location coordinates of the workers, as well as the type and wearing status of safety protective equipment, encodes the detection results into structured data, and generates image feature vectors; S5. Based on the image feature vector and the optimal model parameters, the risk identification model is used to make predictions and obtain the risk assessment results. S3 utilizes a discretized tunicate optimization algorithm to search for the optimal parameter configuration in the parameter space of the risk identification model, including: Define the parameter search space, including: learning rate Batch size and dropout rate ; Construct a discrete index mapping table to map the continuous parameter search space to discrete indexes; Construct the fitness function F: ;in, The accuracy of the model on the validation set. For model complexity, This is the balance coefficient; Perform a discretized optimization process for the tunicate, iteratively optimizing until the termination condition is met, and output the optimal parameter configuration. ; Performing the discretized tunicate optimization process includes: Initialize population data matrix Where N is the population size, and each row represents an individual in a three-dimensional continuous space. Location data in the middle; Update the position data in each iteration: For individual leaders, according to the formula Update continuous location data; For individual followers, according to the formula Calculate new location data; For each continuous value Calculate candidate discrete indices: ; Calculate the mapping distance: ; According to probability Perform random mapping to generate the final discrete index; where, Indicates temperature parameter; Extract actual parameter values ​​from the parameter candidate set based on discrete indexes; Configure the model using the extracted parameter values, evaluate performance on the validation set, calculate fitness values, and update the optimal solution record.

2. The safety detection method for power construction activities according to claim 1, characterized in that: S2, based on the work order risk feature library, train a risk identification model based on multi-task learning, including: Risk feature vectors and historical accident markers are extracted from the work order risk feature database. The dataset is divided into three subsets based on the work type code: live-line work, switching operation, and maintenance work. Based on the data subset, a multi-task learning network architecture is constructed, in which the shared feature extraction layer adopts a three-layer fully connected network with three task output heads: The operation sequence compliance output header is used to predict whether the switching operation steps comply with the standard operating procedures. The safety distance detection output head is used to predict whether the safety distance between the operator and the live equipment meets the requirements of the regulations. The protective equipment specification output head is used to predict whether the use of safety helmets and insulating gloves meets the specification requirements. By setting a multi-task loss function and using the batch gradient descent algorithm for model training, a risk identification model based on multi-task learning is obtained.

3. The safety detection method for power construction activities according to claim 2, characterized in that: The multi-task loss function expression is as follows: ;in, The binary cross-entropy loss is for compliance with the order of operations. For multi-class cross-entropy loss in safe distance detection, To protect against the standard binary cross-entropy loss of the equipment, This represents the task weighting coefficient.

4. The safety detection method for power construction activities according to claim 1, characterized in that: S4 generates image feature vectors, including: Collect real-time image data of the work site and preprocess the images; Identify key targets in images using the YOLOv5 object detection model: Inspection personnel, output bounding box coordinates and confidence level ; Detect safety helmets and insulating gloves, and output the corresponding protective equipment bounding box and confidence level; Detects live equipment and operating equipment, and outputs the bounding box coordinates and category labels of the equipment; The spatial relationship and wearing status of the target are quantitatively analyzed based on the bounding box coordinates to obtain the detection results; The detection results are encoded into structured data to generate image feature vectors.

5. The safety detection method for power construction activities according to claim 4, characterized in that: The spatial relationships and wearing status of the target are quantitatively analyzed based on the bounding box coordinates to obtain the detection results, including: Extract edge sampling point sets from the bounding boxes of the workers and the live equipment, respectively; Calculate the Euclidean distance matrix between point pairs based on the edge sampling point set; The minimum value is extracted from the Euclidean distance matrix and used as the safety distance, which is then used as the result of target spatial relationship detection. The boundary box of the workers is divided into functional areas according to a preset ratio; Calculate the overlap between the boundary frame of the protective equipment and the corresponding functional areas of the human body; Binary wearing status detection results are generated based on the overlap threshold.

6. The safety detection method for power construction activities according to any one of claims 1 to 5, characterized in that: S5, using a risk identification model to make predictions, obtains risk assessment results, including: Configure the parameters of the risk identification model using the optimal model parameters and establish a parameterized model instance; Using image feature vectors as input data, based on a parameterized model instance: The input features are transformed nonlinearly through a shared feature extraction layer; The transformed features are input into multiple task output heads in parallel; Extract the corresponding predicted probability distribution from each output head; Perform threshold judgment and category mapping based on the predicted probability distribution; Based on the threshold judgment and category mapping results, a risk assessment result is generated through weighted aggregation.

7. A safety detection system for power construction activities, used to perform the method described in any one of claims 1 to 6, characterized in that, include: The historical data processing module is used to collect historical power work order data and extract fields such as work content, safety measures, work steps, and hazard pre-control measures from the work orders; it constructs a work order risk feature library based on the extracted fields, and each record in the risk feature library contains a work type code, a risk feature vector, and historical accident markers; The model training module, connected to the historical data processing module, is used to train a risk identification model based on multi-task learning according to the work ticket risk feature library. The risk identification model includes a shared feature extraction layer and multiple task output heads, each of which corresponds to a type of work risk assessment task. The parameter optimization module is used to collect the text data of the current power work order, extract the work content and safety measures fields, and generate a text feature vector; based on... The text feature vectors are used to search for the optimal parameter configuration in the parameter space of the risk identification model using the discretized sea squirt optimization algorithm, so as to obtain the optimal model parameters that match the current work order operation type. The image processing module is used to acquire real-time image data of the work site, perform target detection and feature extraction on the image data, identify the location coordinates of the workers, as well as the type and wearing status of safety protective equipment, encode the detection results into structured data, and generate image feature vectors. The risk assessment module is connected to the model training module, parameter optimization module, and image processing module, respectively. It is used to make predictions based on the image feature vector and the optimal model parameters using the risk identification model to obtain the risk assessment result.

Images