A power construction safety behavior identification and control method and device

Abstract

The application discloses a kind of electric power construction safety behavior identification and control method and device, it is related to electric power safety technical field, the method includes: when carrying out power distribution cabinet electric power construction, personnel image of construction personnel is collected, insulation equipment wearing identification is carried out, obtains insulation equipment wearing parameter distribution;When insulation equipment wearing parameter distribution is qualified, obtain electric power construction task characteristics, and divide power-off area and live area in power distribution cabinet;According to electric power construction task characteristics, the operation trajectory of multiple key positions is predicted, and the operation trajectory sequence set is obtained;According to insulation equipment wearing parameter distribution, operation trajectory sequence set and live area, combine operation experience data, carry out operation trajectory compensation and abnormal prediction, obtain abnormality degree distribution, as identification result and carry out early warning control, solve the problem that prior art lacks abnormal analysis capability and dynamic risk pre-judgment capability, early warning response is low, safety is low, and safety control effect is not good.

Description

Technical Field

[0001] This invention relates to the field of power safety technology, specifically to a method and device for identifying and controlling safe behaviors during power construction. Background Technology

[0002] Electrical construction, especially work near live or partially live equipment such as distribution cabinets, is highly dangerous. Ensuring the safety of construction personnel includes properly wearing personal protective equipment (PPE) before work and maintaining a sufficient safe distance from live equipment during work.

[0003] Traditional safety control methods rely heavily on manual supervision, lacking the ability to analyze anomalies and predict dynamic risks. This leads to the risk of construction workers unintentionally touching or approaching live areas, resulting in low early warning responsiveness, low safety, and ineffective safety control. Therefore, there is a need for a method for identifying and controlling safe behaviors in power construction that can comprehensively assess personnel protective status, predict operational behaviors for risk assessment, and enable proactive intervention. Summary of the Invention

[0004] This application provides a method and device for identifying and controlling safe behaviors during power construction, which addresses the problems in existing technologies that lack the ability to analyze anomalies and predict dynamic risks, resulting in potential hazards such as construction workers unintentionally touching or approaching live areas, leading to low early warning response, low safety, and poor safety control.

[0005] In view of the above problems, this application provides a method and device for identifying and controlling safety behaviors during power construction.

[0006] In a first aspect, this application provides a method for identifying and controlling safety behaviors during power construction, the method comprising: During the electrical construction of the distribution cabinet, images of the construction personnel are collected, and the wearing of insulating equipment is identified to obtain the distribution of insulating equipment wearing parameters. The distribution of insulating equipment wearing parameters includes wearing parameters at multiple key locations. When the distribution of the insulation equipment wearing parameters is qualified, the characteristics of the power construction task are obtained, and the power outage area and the live area are divided in the distribution cabinet; Based on the characteristics of the power construction task, the operation trajectory is predicted at multiple key locations to obtain a set of operation trajectory sequences; Based on the distribution of insulating equipment wearing parameters, the set of work trajectory sequences, and the energized area, combined with the work experience data of construction personnel, work trajectory compensation and anomaly prediction are performed to obtain the anomaly degree distribution, which is used as the identification result for early warning control.

[0007] Secondly, the present invention provides a power construction safety behavior identification and control device, the device comprising: The parameter distribution acquisition module is used to collect images of construction personnel during the power construction of the distribution cabinet, identify the wearing of insulating equipment, and obtain the wearing parameter distribution of the insulating equipment. The wearing parameter distribution of the insulating equipment includes wearing parameters at multiple key locations. The area division module is used to obtain the characteristics of the power construction task when the distribution of the insulation equipment wearing parameters is qualified, and to divide the power outage area and the live area in the distribution cabinet. The work trajectory prediction module is used to predict the work trajectory at multiple key locations based on the characteristics of the power construction task, and obtain a set of work trajectory sequences. The early warning control module is used to perform work trajectory compensation and anomaly prediction based on the distribution of the insulating equipment wearing parameters, the work trajectory sequence set, and the energized area, combined with the work experience data of the construction personnel, to obtain the anomaly degree distribution as the identification result and perform early warning control.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application firstly acquires personnel images in real time during electrical construction at the distribution cabinet and identifies the wearing of insulating equipment, obtaining the distribution of insulating equipment wearing parameters at key locations. This enables the inspection of the personal safety protection status of construction personnel, establishing a reliable personnel safety foundation for subsequent operations. Secondly, when the distribution of insulating equipment wearing parameters is qualified, the power outage area and the energized area are divided within the distribution cabinet based on the characteristics of the electrical construction task, achieving dynamic definition of the risk boundary of the working environment and providing a reference for subsequent behavior prediction and risk assessment. Thirdly, based on the characteristics of the electrical construction task, the operation trajectory is predicted at multiple key locations, obtaining a set of operation trajectory sequences. The prediction model then predicts the operation trajectory sequence, providing a data foundation for risk identification. Finally, by integrating the distribution of insulating equipment wearing parameters, the set of operation trajectory sequences, the energized areas, and the operational experience data of construction personnel, operation trajectory compensation and anomaly prediction are performed to obtain the anomaly degree distribution and use it for early warning and control. This achieves comprehensive risk assessment by integrating multi-source heterogeneous information, and by calculating the anomaly degree distribution, the degree of risk is located, improving the responsiveness of safety management. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a method for identifying and controlling safe behaviors during power construction, as described in this application. Figure 2 This is a structural schematic diagram of a power construction safety behavior identification and control device according to this application.

[0011] In the attached diagram, the components represented by each number are as follows: Parameter distribution acquisition module 11, area division module 12, operation trajectory prediction module 13, early warning control module 14. Detailed Implementation

[0012] This application provides a method and device for identifying and controlling safe behaviors during power construction, which addresses the problems in existing technologies that lack the ability to analyze anomalies and predict dynamic risks, leading to potential hazards such as construction workers unintentionally touching or approaching live areas, resulting in low early warning response, low safety, and poor safety control.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, methods, products, or devices.

[0015] The present invention will now be described in detail with reference to the accompanying drawings.

[0016] Example 1, as Figure 1 As shown, this application provides a method for identifying and controlling safety behaviors during power construction, the method comprising: S10: During the electrical construction of the distribution cabinet, images of the construction personnel are collected, and the wearing of insulating equipment is identified to obtain the distribution of insulating equipment wearing parameters. The distribution of insulating equipment wearing parameters includes wearing parameters at multiple key locations.

[0017] In this embodiment, electrical construction refers to installation, maintenance, debugging, and component replacement operations carried out inside or around the distribution cabinet; personnel images are obtained through image acquisition devices such as cameras, including digital images or video frames of the construction personnel's full body or upper body; insulation equipment wearing recognition is based on computer vision image analysis technology, used to automatically detect and determine whether construction personnel are correctly wearing the necessary insulation protective equipment; insulation equipment wearing parameter distribution is a description of the recognition results, including evaluation values ​​of the wearing status for multiple key positions on the construction personnel's body or equipment.

[0018] Specifically, when construction workers enter the distribution cabinet or designated work area, images are captured by image acquisition equipment. The captured images are then input into an insulation equipment wearing recognition device for analysis. The device locates multiple preset key positions and analyzes and scores the insulation equipment at each position, generating wearing parameters. Ultimately, a distribution of insulation equipment wearing parameters is generated. A higher equipment wearing quality score indicates more qualified wearing.

[0019] Step S10 in the method provided in this application embodiment includes: During the electrical installation of the distribution cabinet, images of the construction workers are captured using the image acquisition equipment on the distribution cabinet. The personnel image is input into the insulating equipment wear recognition device, which identifies and outputs wearing parameters at multiple key locations to obtain the distribution of insulating equipment wearing parameters. The insulating equipment wear recognition device includes multiple attention recognition branches.

[0020] In this embodiment, during the electrical construction of the distribution cabinet, images of the construction workers are first captured using an image acquisition device on the distribution cabinet. The image acquisition device consists of one or more vision sensors fixedly installed on the distribution cabinet body or a supporting structure closely associated with the distribution cabinet.

[0021] Specifically, an image acquisition device, such as an industrial camera, is pre-installed at the center of the top of the cabinet door. When construction workers approach the cabinet to conduct pre-operation safety checks, a full-body color image of the worker is automatically captured. The captured image must clearly show the worker wearing appropriate insulation equipment.

[0022] For example, a person image of a construction worker is captured, clearly showing the worker wearing a safety helmet, insulated gloves, insulated clothing, and insulated boots.

[0023] Secondly, the personnel image is input into the insulating equipment wearing recognition device, which identifies and outputs wearing parameters at multiple key locations to obtain the distribution of insulating equipment wearing parameters. The insulating equipment wearing recognition device includes multiple attention recognition branches.

[0024] Specifically, the insulating equipment wear recognition device is an algorithmic model that performs computer vision recognition tasks. After receiving a person's image, it performs calculations and outputs quantitative or qualitative analysis results on the wearing status of the person's insulating equipment in the image. The attention recognition branch is a neural network substructure inside the recognizer that is inspired by the attention mechanism.

[0025] A person's image is input into an insulating equipment wearing recognition device, which includes multiple attention-based recognition branches. Each device focuses only on a specific body part and uses an attention mechanism to locate the insulating equipment in the target area of ​​the image. Different recognition strategies and decision thresholds are applied to different body parts, and parallel processing improves the accuracy of the recognition. Finally, the outputs of all branches contain the wearing parameters for all key locations, forming an insulating equipment wearing parameter distribution.

[0026] In step S10 of the method provided in this application embodiment, the training step of the insulating equipment wear identifier includes: Based on historical testing data of insulating equipment wearing, a set of sample personnel images was collected, and the wearing parameters of multiple key locations in each sample personnel image were labeled to obtain multiple sets of sample wearing parameters, where each wearing parameter includes a wearing quality score. Based on a convolutional neural network, multiple attention recognition branches are constructed. The input of the multiple attention recognition branches is a person image, and the output is the wearing parameters of multiple key positions. Using the sample personnel image set and multiple sample wearing parameter sets, supervised training and testing are performed on multiple attention recognition branches respectively. After the test accuracy is qualified, the training of the insulating equipment wearing recognition device is completed.

[0027] In this embodiment of the application, firstly, based on historical detection data of insulating equipment wearing, a set of sample personnel images is collected, and the wearing parameters of multiple key positions in each sample personnel image are labeled to obtain multiple sets of sample wearing parameters, wherein each wearing parameter includes a wearing quality score.

[0028] Specifically, historical inspection data is a database of original records related to the wearing of insulating equipment during past power construction safety supervision processes; the sample personnel image set refers to a diverse image dataset constructed for training; annotation is the operation of constructing supervised learning datasets in machine learning, in which annotators analyze the sample personnel images according to clear specifications, identify multiple preset key locations in the images, and provide an evaluation of the wearing status of insulating equipment at each location; the wearing quality score is a quantitative form of the annotation results, used to accurately describe the standardization and safety of the wearing of insulating equipment at a certain key location.

[0029] Specifically, historical monitoring data from a past period was retrieved, and frontal and side images of construction workers were captured under different lighting and angles to form a sample personnel image set. Each personnel image was then labeled. For each key position (head), the standard of the insulating equipment worn was assessed, and a corresponding wearing quality score was given based on the specific wearing condition of the insulating equipment. After all images were labeled, a sample wearing parameter set was formed.

[0030] Secondly, based on a convolutional neural network, multiple attention recognition branches are constructed. The input to these branches is a person image, and the output is the wearing parameters at multiple key locations. The convolutional neural network is a deep learning model specifically designed for processing grid-like data.

[0031] Specifically, based on a convolutional neural network, parallel attention recognition branches are constructed to identify different parts of the construction worker's body. Each branch has a similar structure, analyzing the person's image to generate an attention weight map that highlights the head area and dims other areas, obtaining focused features. These focused features are then processed through several dedicated convolutional and fully connected layers, ultimately outputting a wear quality score representing the corresponding identified part of the branch. Given an input image of a person, the prediction and recognition model will simultaneously output wear parameters for multiple key locations.

[0032] Finally, using a sample set of personnel images and multiple sets of sample wearing parameters, supervised training and testing were conducted on multiple attention recognition branches. After achieving satisfactory test accuracy, the training of the insulating equipment wearing recognizer was completed. Supervised training is the most important training method in machine learning; achieving satisfactory test accuracy means that the model's performance metrics on the test set meet pre-defined business requirements.

[0033] For example, the steps to construct multiple attention branches using a convolutional neural network are as follows: Model Structure: Each attention branch has the same structure, but the key recognition areas differ. All branches employ a convolutional neural network (CNN) structure, which mainly includes: an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. The input layer receives the raw image data; the convolutional layers extract features using convolutional kernels to generate feature maps; the pooling layers downsample the feature maps to reduce image size; the fully connected layers flatten the extracted features and map them to the output space; and the output layer outputs the prediction results based on the task type.

[0034] Model Training: Sample images are divided into training, validation, and test sets in a 7:2:1 ratio. Using images of people as input and clothing parameters at multiple key locations as output, supervised training is performed on each attention branch. In each training round, the model reads images of people, and each attention branch identifies specific key areas. Each batch of images is forward-propagated through the model, and each attention branch outputs a predicted score for its assigned area. Then, mean squared error (MSE) loss is used to calculate the loss between the predicted score of each branch and its ground truth value. Backpropagation is used to calculate the gradient of the total loss with respect to all model parameters, and the Adam optimizer is used to update the parameters to make the model's output score closer to the true value. This process is repeated until the model converges.

[0035] Simultaneously, an independent validation set is used to monitor model performance and prevent overfitting. After training, the model is evaluated using a test set that was never used in training, and the accuracy of each attention recognition branch on the test set is calculated. When the test accuracy of all branches is higher than 95%, the training of the insulating equipment wear recognizer is considered complete, and it is packaged and deployed to the production environment for real-time recognition tasks.

[0036] In this embodiment, an image acquisition device fixed to the power distribution cabinet ensures the stability of the image acquisition perspective and environment, providing high-quality, standardized input for subsequent recognition. By constructing a recognizer with multiple attention recognition branches, and using the input personnel image and wearing parameters of key locations for each branch, the wear status of insulating equipment on multiple body parts is scored, reducing the impact of background interference and mutual occlusion between components, and improving the accuracy of recognition in complex field environments. Simultaneously, the generalization ability of the model is ensured.

[0037] S20: When the distribution of the insulation equipment wearing parameters is qualified, obtain the characteristics of the power construction task and divide the power outage area and the live area in the distribution cabinet; In this embodiment of the application, the characteristics of the power construction task are the attribute information of the construction task; the power outage area is the physical space within the distribution cabinet where power has been confirmed to be out of power, tested for electricity, and safety measures have been taken, allowing personnel to access and work; the energized area is the physical space within the distribution cabinet where power has not been cut off or where, although power has been cut off, sufficient safety measures have not been taken, it is still considered to be energized, and personnel are prohibited from touching it.

[0038] Specifically, after confirming that personnel protection is adequate, the distribution of insulation equipment wearing parameters is assessed to ensure it is within acceptable limits. If it is not, the process may be interrupted and an alarm issued, requiring personnel to rectify the issue. If it is within acceptable limits, the process continues. Next, the characteristics of the power construction task are obtained, and combined with the electrical wiring diagram and current operating status of the distribution cabinet, the power outage area and the energized area are divided on the three-dimensional spatial model or two-dimensional layout diagram of the distribution cabinet.

[0039] Step S20 in the method provided in this application embodiment includes: Determine whether all wearing parameters within the distribution of wearing parameters of the insulating equipment meet the wearing parameter threshold. If yes, the distribution of wearing parameters of the insulating equipment is qualified; otherwise, it is unqualified and a prompt is given. When the distribution of the insulation equipment wearing parameters is qualified, the characteristics of the power construction task are obtained; Input the characteristics of the power construction task into the power outage area classification table, and output the power outage areas. The area outside the power outage zone within the distribution cabinet is demarcated to obtain the energized area.

[0040] In this embodiment, it is first determined whether all wearing parameters within the insulation equipment wearing parameter distribution meet the wearing parameter threshold. If so, the insulation equipment wearing parameter distribution is qualified; otherwise, it is unqualified, and a prompt is issued. Here, "all wearing parameters" refers to each specific wearing parameter value contained in the insulation equipment wearing parameter distribution data structure; the wearing parameter threshold is a critical value used to determine whether the insulation equipment wearing at a certain key position is qualified; and "qualified / unqualified" is the binary output result of the judgment action.

[0041] Specifically, all wearing parameters within the insulation equipment wearing parameter distribution are compared with wearing parameter thresholds. If every wearing parameter in the distribution meets the condition of being greater than or equal to the wearing parameter threshold, the insulation equipment wearing parameter distribution is considered qualified. If any wearing parameter does not meet the condition of being greater than or equal to the wearing parameter threshold, the insulation equipment wearing parameter distribution is considered unqualified. For unqualified insulation equipment wearing parameter distributions, construction personnel can be reminded through voice prompts or by disabling operation permissions to ensure personnel safety. The wearing parameter threshold is set according to the wearing standards that ensure the safety of construction personnel after wearing the equipment. If the wearing parameter threshold is met, work can be carried out in the normal work space, ensuring the safety of construction personnel. The wearing parameter threshold should not be set too high or too low. If the wearing parameter threshold is too low, it will cause unnecessary interference to normal operations; if the wearing parameter threshold is too high, it will cause problems with untimely safety warnings, affecting the safety of construction tasks. For example, if the wearing parameter range is 0-100 points, the wearing parameter threshold can be set to 80 points.

[0042] Secondly, when the distribution of insulation equipment wearing parameters is within acceptable limits, the characteristics of the power construction task are acquired. These characteristics are key information data representing the specific attributes of the construction task. Specifically, if the distribution of insulation equipment wearing parameters is deemed acceptable, the actions of construction personnel will not be restricted, and the characteristics of the power construction task will be acquired promptly from the task terminal.

[0043] For example, after being deemed qualified, the characteristics of the power construction task of the construction personnel are obtained through the task terminal, which may include task type, equipment number, equipment voltage, construction scope, safety measures, etc., to facilitate the safety handling of specific areas during subsequent construction operations.

[0044] Next, the characteristics of the power construction task are input into the power outage area classification table, and the power outage areas are output. The power outage area classification table is a lookup table that stores the correspondence between the characteristics of different types of power construction tasks and the corresponding standard power outage areas in specific types of distribution cabinets. Under the division of power outage tasks, the safe contact range is clearly defined in the three-dimensional spatial model of the distribution cabinet according to safety regulations and implemented safety measures.

[0045] Specifically, the characteristics of the acquired power construction tasks are input into a power outage area classification table. The standard power outage areas corresponding to the power construction tasks are queried, and the resulting spatial range is used as the power outage area for the operation. During construction, workers can identify the power outage areas based on the specific power construction task's power outage standards and implement power outage measures within these areas to ensure the completion of the construction work and the safety of the workers.

[0046] Finally, the area outside the power outage zone within the distribution cabinet is demarcated to obtain the energized area. Specifically, the area outside the power outage zone within the distribution cabinet refers to all space outside the power outage zone in the internal 3D spatial model of the distribution cabinet; the energized area is all space outside the power outage zone, posing a risk of electric shock. In detail, after obtaining the power outage zone, a standard 3D model is used to remove the power outage zone from the complete cabinet space model, and the remaining space is designated as the energized area for subsequent comparison.

[0047] In step S20 of the method provided in this application embodiment, the power outage area classification table includes a mapping relationship between the sample power construction task feature set and the sample power outage area set, and the power construction task features include the power construction range.

[0048] In this embodiment, the power outage area classification table includes a mapping relationship between a sample set of power construction task features and a sample set of power outage areas. The power construction task features include the power construction scope. Specifically, by obtaining a sample set of power construction task features and a sample set of power outage areas from a past historical period, a mapping relationship can be constructed between these two sets using a mapping function to obtain the power outage area classification table. This table maps the power construction task features to power outage areas for subsequent power outage area division. The power construction scope refers to the specific spatial location or equipment components involved during construction, facilitating the lookup of power outage areas in the power outage area classification table.

[0049] In this embodiment, safety access checks are transformed into automatically executable quantitative decisions through wearable parameter threshold judgment rules, and immediate prompts are issued when non-compliance occurs, ensuring construction safety. Simultaneously, a power outage area classification table is used to query, match, and output power outage areas, automatically and accurately determining safe working boundaries based on specific construction task characteristics. The power outage area classification table, constructed based on the mapping relationship between historical tasks and areas, ensures the standardization and consistency of area division. Finally, based on the power outage areas, energized areas are divided into sets, completing the construction of the work environment risk space and providing a stable prediction space for subsequent dynamic behavior risk prediction.

[0050] S30: Based on the characteristics of the power construction task, predict the operation trajectory at multiple key locations to obtain a set of operation trajectory sequences; In this embodiment, the key location is a key point or feature point on the body or tool of the construction worker that needs to be tracked, which directly reflects the work behavior; the work trajectory prediction is based on the current task information and uses a model to predict the movement path of the construction worker in the three-dimensional space of the key location in the future.

[0051] Specifically, before the operation, the work trajectory predictor predicts the characteristics of the power construction task and personnel, and outputs a sequence of work trajectories for multiple preset key locations during the operation.

[0052] Step S30 in the method provided in this application embodiment includes: A work trajectory predictor is obtained, wherein the work trajectory predictor includes multiple work trajectory prediction branches corresponding to multiple key locations, wherein the input data of each work trajectory prediction branch is personnel characteristics and power construction task characteristics, and the output data is a sequence of work trajectories for key locations, and each work trajectory includes the spatial three-dimensional coordinates of the corresponding key location within the work time range. The characteristics of the construction personnel are obtained and combined with the characteristics of the power construction task, which are then input into the work trajectory predictor to predict the work trajectory sequence of multiple key locations, thus obtaining a set of work trajectory sequences of multiple key locations.

[0053] In this embodiment of the application, a work trajectory predictor is first obtained. The work trajectory predictor includes multiple work trajectory prediction branches corresponding to multiple key locations. The input data of each work trajectory prediction branch are personnel characteristics and power construction task characteristics, and the output data is the work trajectory sequence of key locations. Each work trajectory includes the spatial three-dimensional coordinates of the corresponding key location within the work time range.

[0054] The work trajectory predictor is a software model or algorithm module used to predict the dynamic process of construction operations; multiple key locations are the body nodes of the construction personnel; multiple work trajectory prediction branches are independent sub-models or sub-networks of the predictor's internal architecture; personnel characteristics are data describing the individual attributes of the construction personnel. The operation time range is the duration of the entire operation process; the three-dimensional spatial coordinates are the instantaneous positions of the key locations in the three-dimensional space inside the distribution cabinet.

[0055] Specifically, the process begins by acquiring a work trajectory predictor, which is then used to predict the work trajectories of construction workers at multiple key locations. Given the trained work trajectory predictor, if personnel characteristics and power construction task characteristics are input, the predictor will activate the corresponding work trajectory prediction branches for each key location, performing parallel predictions, processing the movement patterns of different locations, and outputting the work trajectory sequences for each key location.

[0056] For example, using personnel characteristics and power construction task characteristics as input data, an LSTM model is used to construct a work trajectory predictor. Specifically, for N key locations corresponding to construction personnel, N work trajectory prediction branches are constructed, with each branch identifying one key location. Long Short-Term Memory (LSTM) networks are a special type of RNN used to solve the vanishing gradient problem, capture and understand complex dependencies in long sequences, and are widely used in many sequence learning tasks such as speech recognition, machine translation, and time series analysis.

[0057] The steps for constructing a working trajectory predictor using a Long Short-Term Memory (LSTM) network are as follows: Model data: The personnel characteristics and power construction task characteristics are deeply processed. The parameters are normalized to the range of [-1,1] or [0,1] through standardization to eliminate the difference in dimensions. The sliding window technique is used to divide the personnel characteristics and power construction task characteristics into corresponding input pairs, which are used as training samples for the LSTM model.

[0058] Model Structure: The model comprises multiple parallel LSTM branches for job trajectory prediction, each corresponding to a key location. Each job trajectory prediction branch is an LSTM network, with all LSTM units receiving the same input. Each job trajectory prediction branch consists of an input layer, hidden layers, fully connected layers, and an output layer. The input layer receives personnel features and power construction task features. The hidden layer, located between the input and output layers, extracts abstract features from the batch size, time steps, and feature dimensions of the personnel and power construction task feature input pairs. The fully connected layer uses the ReLU activation function to integrate and transform the features extracted by the hidden layer. The output layer employs a linear activation function to output the job trajectory sequence for each key location.

[0059] Model Training: The model is divided into training, validation, and test sets in a 7:2:1 ratio. During training, the model is input, and each LSTM branch is activated. The Adam optimizer is used for parameter updates, with an initial learning rate of 0.001 and a learning rate decay strategy. Mean squared error is used as the loss function, and the sum or weighted sum of the losses of each branch is used as the total loss. The model parameters are optimized using the time backpropagation algorithm. Performance is monitored on the validation set during training to prevent overfitting. The model performance is evaluated on the validation set after each training epoch. Training is terminated when the validation set loss no longer decreases for 10 consecutive epochs. After training, a final evaluation is performed on the test set. If the mean absolute percentage error of the load prediction is below 2.5%, the corresponding N job trajectory prediction branches, i.e., the job trajectory predictor, are obtained.

[0060] Secondly, the characteristics of the construction personnel are obtained and combined with the characteristics of the power construction task, which are then input into the work trajectory predictor to predict the work trajectory sequences of multiple key locations, thus obtaining a set of work trajectory sequences of multiple key locations.

[0061] Specifically, based on the input personnel characteristics and power construction task characteristics, the correlation between the key locations of historical construction personnel and the characteristics of power construction tasks is used to analyze the operational behavior of specific construction personnel and the characteristics of power construction tasks. The LSTM of each branch of the model generates a complete sequence of predicted operation trajectories. Finally, the outputs of all branches are collected to form a set of operation trajectory sequences for subsequent risk prediction.

[0062] For example, the personnel characteristics of construction workers (such as height, arm length, etc.) are obtained from a personnel database. These personnel characteristics are then combined with the characteristics of the power construction task and input into a work trajectory predictor. After calculation, the predictor outputs the work trajectory sequence for specific location 1 (branch A), the work trajectory sequence for specific location 2 (branch B), and the work trajectory sequence for specific location 3 (branch C). Finally, these are integrated to obtain the work trajectory sequence.

[0063] In this embodiment, a work trajectory predictor with multiple branches predicts the future movement trajectory of different key parts in parallel, comprehensively considering the specific tasks and the individualization of personnel, making the prediction results more consistent with the actual work scenario. Simultaneously, it provides a data foundation for the spatial relationship between the work trajectory sequence and the energized area, and for risk detection.

[0064] S40: Based on the distribution of the insulating equipment wearing parameters, the set of work trajectory sequences, and the energized area, combined with the work experience data of the construction personnel, work trajectory compensation and anomaly prediction are performed to obtain the anomaly degree distribution, which is used as the identification result for early warning control.

[0065] In this embodiment, the work experience data is the historical work performance record of the construction personnel, which is used to quantify their experience level and behavioral reliability; the work trajectory compensation takes into account that the prediction cannot be absolutely accurate, and the actual actions of different personnel will fluctuate around the predicted trajectory; the anomaly prediction is to perform real-time or forward-looking collision detection and distance analysis between the compensated work trajectory range and the spatial model of the electrified area.

[0066] Specifically, the process begins by acquiring the operational experience data of construction workers. Then, combining this with the distribution of insulating equipment wearing parameters, trajectory compensation is performed on the operational trajectory sequence set to generate a range sequence set of operational trajectories for the construction workers. Next, the proportion of trajectories whose distance from the energized area is less than a threshold is calculated as the anomaly score. Finally, the anomaly scores of the construction workers at multiple key locations are calculated, forming an anomaly score distribution.

[0067] Step S40 in the method provided in this application embodiment includes: Obtain work experience data from construction workers, including the percentage of abnormal work incidents. Based on the distribution of the insulating equipment wearing parameters and operational experience data, trajectory compensation is performed on the operational trajectory sequence set to obtain the operational trajectory range sequence set; Based on the set of operation trajectory range sequences and the energized area, anomaly prediction is performed to obtain the anomaly degree distribution; The anomaly distribution is used as the identification result for early warning control.

[0068] In this embodiment, the work experience data of construction workers is first obtained, including the proportion of abnormal work occurrences. The work experience data is a set of data on the historical work performance, skill level, and safe work habits of the construction workers; the proportion of abnormal work occurrences is the ratio of the number of times a construction worker was recorded as performing unsafe work to their total number of work occurrences.

[0069] Specifically, the system retrieves construction workers' work experience data over a past period by querying the backend database. If a violation is found in the work experience data, it is recorded as an abnormal operation. The abnormal operation frequency is calculated as the ratio of abnormal operations to the total number of operations. This abnormal operation frequency reflects the construction workers' safety awareness. A low abnormal operation frequency indicates weak safety awareness and a higher likelihood of accidents; conversely, a high frequency indicates strong safety awareness and a higher level of safety.

[0070] Secondly, based on the distribution of insulating equipment wearing parameters and operational experience data, trajectory compensation is performed on the operational trajectory sequence set to obtain the operational trajectory range sequence set. Trajectory compensation involves correcting the operational trajectory sequence set by estimating the fluctuation range; the operational trajectory range sequence set is the three-dimensional coordinate interval range obtained after trajectory compensation.

[0071] Specifically, based on the distribution of insulation equipment wearing parameters and operational experience data, a personalized compensation value is calculated and determined to adjust the operational trajectory sequence set.

[0072] Next, based on the sequence set of work trajectory ranges and the energized area, anomaly prediction is performed to obtain the anomaly degree distribution. Anomaly prediction involves calculating the risk probability based on a spatial model of the compensated predicted trajectory range and known hazardous areas before the actual operation occurs. The anomaly degree distribution represents the risk level of the movement trajectory of critical components encroaching upon or excessively approaching the energized area within the predicted operation timeframe.

[0073] Specifically, the set of operation trajectory range sequences is compared with the energized area, the proportion of key location trajectory ranges that are less than the safe distance threshold of the energized area is calculated, the anomaly degree is calculated, and anomaly degree analysis is performed on all operation trajectory range sequences.

[0074] Finally, the anomaly distribution is used as the identification result for early warning control. The identification result is the final output of the entire safety behavior identification method. Early warning control is a proactive safety intervention measure triggered based on the identification result. Specifically, after anomaly determination, early warning control is triggered based on the anomaly value at each key location in the anomaly distribution. If any anomaly exceeds the safe distance threshold, a strong early warning can be issued before the risk time point through audible and visual alarms and voice prompts, and the cabinet door can be prevented from opening or the operating power supply can be suspended.

[0075] In step S40 of the method provided in this application embodiment, trajectory compensation is performed on the work trajectory sequence set based on the distribution of the insulating equipment wearing parameters and operational experience data to obtain the work trajectory range sequence set, including: Obtain the average percentage of abnormal operations in power construction, calculate the ratio of the percentage of abnormal operations in the operational experience data to the average percentage of abnormal operations, and obtain the experience compensation coefficient. Obtain the baseline wearing parameters, calculate the mean of the distribution of the wearing parameters of the insulating equipment, obtain the average wearing parameters of the construction personnel, and calculate the ratio of the baseline wearing parameters to the average wearing parameters to obtain the wearing compensation coefficient. A preset trajectory error scale is obtained, and the preset trajectory error scale is adjusted and calculated using the empirical compensation coefficient and the wearable compensation coefficient to obtain the adjusted trajectory error scale. The trajectory error scale is adjusted to compensate for the trajectory error of the operation trajectory sequence set to obtain the operation trajectory range sequence set. Each operation trajectory range includes the spatial three-dimensional coordinate interval of the corresponding key position within the operation time range.

[0076] In this embodiment, the average proportion of abnormal operations in power construction is first obtained, and the ratio of the proportion of abnormal operations within the operational experience data to the average proportion of abnormal operations is calculated to obtain the experience compensation coefficient. The average proportion of abnormal operations is the arithmetic mean or median of the proportions of abnormal operations for all construction workers within a certain range and a certain statistical period, representing the average level of overall safety performance over the historical period.

[0077] Specifically, the experience compensation coefficient represents the degree to which the historical behavioral risk of current construction workers deviates from the group average level. Experience compensation coefficient = (percentage of abnormal operations) / (average percentage of abnormal operations). If the experience compensation coefficient is greater than 1, it indicates that the worker's historical risk is higher than the average level, requiring increased compensation; if the experience compensation coefficient is less than 1, it indicates that their risk is lower than the average level, and compensation can be appropriately reduced.

[0078] For example, the average number of abnormal operations is 0.05, the number of abnormal operations by construction workers is 0.02, and the experience compensation coefficient is 0.02 / 0.05 = 0.4.

[0079] Secondly, the baseline wearing parameters are obtained, the mean of the distribution of insulation equipment wearing parameters is calculated, the average wearing parameters of construction workers are obtained, and the ratio of the baseline wearing parameters to the average wearing parameters is calculated to obtain the wearing compensation coefficient. The baseline wearing parameters represent the ideal wearing quality scoring standard; the average wearing parameters are the arithmetic mean of the wearing parameters at multiple key locations in the insulation equipment wearing parameter distribution, representing the average level of the overall wearing quality of the construction workers.

[0080] Specifically, the average wearing parameter is the average of the sum of the wearing parameters at all key positions. Therefore, the wearing compensation coefficient = baseline wearing parameter / average wearing parameter. The larger the wearing compensation coefficient, the less ideal the wearing is, and the more compensation is needed.

[0081] For example, the preset baseline wearing parameters are 100 points, and the current distribution of insulation equipment wearing parameters for construction workers is [critical position 1:95, critical position 1:90, critical position 1:92, critical position 1:85, critical position 1:100]. The average wearing parameters are (95+90+92+85+100) / 5=92.4 points, and the wearing compensation coefficient is 100 / 92.4≈1.08.

[0082] Next, a preset trajectory error scale is obtained. Using empirical compensation coefficients and wearable compensation coefficients, the preset trajectory error scale is adjusted and calculated to obtain the adjusted trajectory error scale. The preset trajectory error scale is the basic trajectory compensation radius value, representing the positional deviation distance that occurs during the operation; the adjusted trajectory error scale is the trajectory compensation radius obtained after adjustment.

[0083] Specifically, the adjusted trajectory error scale is obtained by multiplying the preset trajectory error scale by the average of the empirical compensation coefficient and the wearing compensation coefficient. That is, the adjusted trajectory error scale = preset trajectory error scale × empirical compensation coefficient × wearing compensation coefficient. Among them, the more times the construction personnel have had abnormal operations in the past, the lower the wearing quality score, and the larger the trajectory error scale.

[0084] For example, assuming the preset trajectory error scale is 0.05m, the adjusted trajectory error scale is approximately 0.05 × 0.4 × 1.08 = 0.02m.

[0085] Finally, by adjusting the trajectory error scale, trajectory error compensation is performed on the set of work trajectory sequences to obtain a set of work trajectory range sequences. Each work trajectory range includes the spatial three-dimensional coordinate interval of the corresponding key position within the work time range. The spatial three-dimensional coordinate interval represents all possible spatial positions of a body part or tool at a given moment.

[0086] Specifically, the calculated adjustment trajectory error scale n is used to compensate the trajectory sequence of the operation trajectory sequence set. The (x,y,z) of one of the predicted points in the sequence is compensated to obtain (x±n,y±n,z±n), generating a spherical interval. This process is repeated for all predicted points to obtain the operation trajectory range sequence set.

[0087] For example, if the trajectory error scale is adjusted to 0.02 meters, then the compensation for (x,y,z) of one of the predicted points is (x±0.02,y±0.02,z±0.02).

[0088] In step S40 of the method provided in this application embodiment, anomaly prediction is performed based on the work trajectory range sequence set and the energized area to obtain anomaly degree distribution, including: Based on the set of operation trajectory range sequences and the energized area, the operation trajectory ranges that are less than the safe distance threshold from the energized area are extracted, and the proportion is calculated to obtain multiple anomalies. By aggregating multiple anomalies at multiple key locations, the anomaly distribution is obtained.

[0089] In this embodiment, firstly, based on the work trajectory range sequence set and the energized area, work trajectory ranges whose distance from the energized area is less than a safe distance threshold are extracted, and their proportions are calculated to obtain multiple anomaly degrees. Specifically, the minimum distance between the edge of each work trajectory range and the energized area is calculated, the proportion of work trajectory ranges whose distance from the energized area is less than the safe distance threshold is extracted, and the proportion of work trajectory ranges whose distance from the energized area is less than the safe distance threshold to the distances of all energized areas is used as the anomaly degree, thus obtaining multiple anomaly degrees between the edge of the work trajectory range and the energized area.

[0090] Secondly, multiple anomalies from multiple key locations are aggregated to obtain the anomaly distribution. Specifically, multiple anomalies from key locations of construction workers are integrated to obtain the complete anomaly analysis results for construction workers.

[0091] In this embodiment, the error range of the preliminary predicted trajectory is adjusted using operational experience data and insulation equipment wearing parameters. The preset error scale is dynamically adjusted using experience compensation coefficients and wearing compensation coefficients to improve the accuracy of risk assessment. A sequence set of operational trajectory ranges is generated, enhancing the realism and adaptability of the risk prediction model. The anomaly degree of each location is obtained by calculating the proportion of time the trajectory range is less than the safety threshold relative to the energized area. All anomalies are aggregated into an anomaly degree distribution, which is then combined with spatial safety distance requirements and temporal probability statistics to obtain the anomaly degree distribution. Finally, the anomaly degree distribution serves as the identification result, driving early warning control and enabling timely safety intervention, thereby improving the safety of power construction safety management.

[0092] The embodiments of this application, through the above specific implementation methods, achieve the following technical effects: In this embodiment, firstly, by limiting the use of image acquisition equipment fixed on the power distribution cabinet, the stability of the image acquisition perspective and environment is ensured, providing high-quality and standardized input for subsequent recognition. Then, by constructing a recognizer containing multiple attention recognition branches, and using the input personnel image and wearing parameters of key locations for each branch, the wearing status of insulating equipment on multiple body parts is scored, reducing the impact of background interference and mutual occlusion between components, and improving the accuracy of recognition in complex on-site environments. Simultaneously, the generalization ability of the model is ensured.

[0093] Secondly, by using wearable parameter threshold judgment rules, safety access checks are transformed into automatically executable quantitative decisions, with immediate alerts for non-compliance to ensure construction safety. Simultaneously, a power outage area classification table is used to query, match, and output power outage areas, automatically and accurately determining safe work boundaries based on specific construction task characteristics. The power outage area classification table, built based on the mapping relationship between historical tasks and areas, ensures the standardization and consistency of area division. Finally, based on power outage areas, energized areas are divided into sets, completing the construction of the work environment risk space and providing a stable prediction space for subsequent dynamic behavior risk prediction.

[0094] Furthermore, by employing a work trajectory predictor with multiple branches, future movement trajectories for different key components are predicted in parallel. This approach comprehensively considers the specific tasks and individual personnel needs, making the prediction results more closely aligned with actual operational scenarios. Simultaneously, it provides a data foundation for the spatial relationship between work trajectory sequences and energized areas, as well as for risk detection.

[0095] Finally, by using operational experience data and insulation equipment wearing parameters, the error range of the initial predicted trajectory is adjusted. The preset error scale is dynamically adjusted using experience compensation coefficients and wearing compensation coefficients to improve the accuracy of risk assessment. A sequence set of operational trajectory ranges is generated, enhancing the realism and adaptability of the risk prediction model. The anomaly degree of each location is obtained by calculating the proportion of time the trajectory range is less than the safety threshold relative to the energized area. All anomalies are aggregated into an anomaly degree distribution. Combining spatial safety distance requirements with temporal probability statistics yields the anomaly degree distribution. Ultimately, the anomaly degree distribution serves as the identification result, driving early warning control and enabling timely safety interventions, thus improving the safety of power construction safety management.

[0096] Example 2, as Figure 2 As shown, based on the same inventive concept as the power construction safety behavior identification and control method provided in Embodiment 1, this embodiment of the invention also provides a power construction safety behavior identification and control device, including: The parameter distribution acquisition module 11 is used to collect images of construction personnel during the power construction of the distribution cabinet, identify the wearing of insulating equipment, and obtain the wearing parameter distribution of insulating equipment, wherein the wearing parameter distribution of insulating equipment includes wearing parameters of multiple key locations. The area division module 12 is used to obtain the characteristics of the power construction task when the distribution of the insulation equipment wearing parameters is qualified, and to divide the power outage area and the live area in the distribution cabinet. The operation trajectory prediction module 13 is used to predict the operation trajectory of multiple key locations based on the characteristics of the power construction task, and obtain an operation trajectory sequence set. The early warning control module 14 is used to perform work trajectory compensation and anomaly prediction based on the distribution of the insulating equipment wearing parameters, the work trajectory sequence set and the energized area, combined with the work experience data of the construction personnel, to obtain the anomaly degree distribution as the identification result and perform early warning control.

[0097] In one embodiment, the parameter distribution acquisition module 11 is used for: During the electrical construction of the distribution cabinet, images of the construction personnel are collected, and the wearing of insulating equipment is identified to obtain the distribution of insulating equipment wearing parameters. The distribution of insulating equipment wearing parameters includes wearing parameters at multiple key locations. When the distribution of the insulation equipment wearing parameters is qualified, the characteristics of the power construction task are obtained, and the power outage area and the live area are divided in the distribution cabinet; Based on the characteristics of the power construction task, the operation trajectory is predicted at multiple key locations to obtain a set of operation trajectory sequences; Based on the distribution of insulating equipment wearing parameters, the set of work trajectory sequences, and the energized area, combined with the work experience data of construction personnel, work trajectory compensation and anomaly prediction are performed to obtain the anomaly degree distribution, which is used as the identification result for early warning control.

[0098] During the electrical installation of the distribution cabinet, images of the construction workers are collected, and their insulation equipment is identified to obtain the distribution of insulation equipment wearing parameters, including: During the electrical installation of the distribution cabinet, images of the construction workers are captured using the image acquisition equipment on the distribution cabinet. The personnel image is input into the insulating equipment wear recognition device, which identifies and outputs wearing parameters at multiple key locations to obtain the distribution of insulating equipment wearing parameters. The insulating equipment wear recognition device includes multiple attention recognition branches.

[0099] The training steps for the insulating equipment wear identifier include: Based on historical testing data of insulating equipment wearing, a set of sample personnel images was collected, and the wearing parameters of multiple key locations in each sample personnel image were labeled to obtain multiple sets of sample wearing parameters, where each wearing parameter includes a wearing quality score. Based on a convolutional neural network, multiple attention recognition branches are constructed. The input of the multiple attention recognition branches is a person image, and the output is the wearing parameters of multiple key positions. Using the sample personnel image set and multiple sample wearing parameter sets, supervised training and testing are performed on multiple attention recognition branches respectively. After the test accuracy is qualified, the training of the insulating equipment wearing recognition device is completed.

[0100] In one embodiment, the region division module 12 is used for: Determine whether all wearing parameters within the distribution of wearing parameters of the insulating equipment meet the wearing parameter threshold. If yes, the distribution of wearing parameters of the insulating equipment is qualified; otherwise, it is unqualified and a prompt is given. When the distribution of the insulation equipment wearing parameters is qualified, the characteristics of the power construction task are obtained; Input the characteristics of the power construction task into the power outage area classification table, and output the power outage areas. The area outside the power outage zone within the distribution cabinet is demarcated to obtain the energized area.

[0101] The power outage area classification table includes a mapping relationship between the sample power construction task feature set and the sample power outage area set. The power construction task features include the power construction scope.

[0102] In one embodiment, the job trajectory prediction module 13 is used for: A work trajectory predictor is obtained, wherein the work trajectory predictor includes multiple work trajectory prediction branches corresponding to multiple key locations, wherein the input data of each work trajectory prediction branch is personnel characteristics and power construction task characteristics, and the output data is a sequence of work trajectories for key locations, and each work trajectory includes the spatial three-dimensional coordinates of the corresponding key location within the work time range. The characteristics of the construction personnel are obtained and combined with the characteristics of the power construction task, which are then input into the work trajectory predictor to predict the work trajectory sequence of multiple key locations, thus obtaining a set of work trajectory sequences of multiple key locations.

[0103] In one embodiment, the early warning control module 14 is used for: Obtain work experience data from construction workers, including the percentage of abnormal work incidents. Based on the distribution of the insulating equipment wearing parameters and operational experience data, trajectory compensation is performed on the operational trajectory sequence set to obtain the operational trajectory range sequence set; Based on the set of operation trajectory range sequences and the energized area, anomaly prediction is performed to obtain the anomaly degree distribution; The anomaly distribution is used as the identification result for early warning control.

[0104] Specifically, based on the distribution of insulating equipment wearing parameters and operational experience data, trajectory compensation is performed on the operational trajectory sequence set to obtain an operational trajectory range sequence set, including: Obtain the average percentage of abnormal operations in power construction, calculate the ratio of the percentage of abnormal operations in the operational experience data to the average percentage of abnormal operations, and obtain the experience compensation coefficient. Obtain the baseline wearing parameters, calculate the mean of the distribution of the wearing parameters of the insulating equipment, obtain the average wearing parameters of the construction personnel, and calculate the ratio of the baseline wearing parameters to the average wearing parameters to obtain the wearing compensation coefficient. A preset trajectory error scale is obtained, and the preset trajectory error scale is adjusted and calculated using the empirical compensation coefficient and the wearable compensation coefficient to obtain the adjusted trajectory error scale. The trajectory error scale is adjusted to compensate for the trajectory error of the operation trajectory sequence set to obtain the operation trajectory range sequence set. Each operation trajectory range includes the spatial three-dimensional coordinate interval of the corresponding key position within the operation time range.

[0105] Among them, based on the set of operation trajectory range sequences and the energized area, anomaly prediction is performed to obtain the anomaly degree distribution, including: Based on the set of operation trajectory range sequences and the energized area, the operation trajectory ranges that are less than the safe distance threshold from the energized area are extracted, and the proportion is calculated to obtain multiple anomalies. By aggregating multiple anomalies at multiple key locations, the anomaly distribution is obtained.

[0106] The embodiments of this application, through the above specific implementation methods, achieve the following technical effects: In the proposed embodiment, firstly, by limiting the use of image acquisition equipment fixed to the power distribution cabinet, the stability of the image acquisition perspective and environment is ensured, providing high-quality and standardized input for subsequent recognition. Then, by constructing a recognizer containing multiple attention recognition branches, and using the input personnel image and wearing parameters for each key location for each branch, the wearing status of insulating equipment on multiple body parts is scored, reducing the impact of background interference and mutual occlusion between components, thus improving the accuracy of recognition in complex on-site environments. Simultaneously, the generalization ability of the model is ensured.

[0107] Secondly, by using wearable parameter threshold judgment rules, safety access checks are transformed into automatically executable quantitative decisions, with immediate alerts for non-compliance to ensure construction safety. Simultaneously, a power outage area classification table is used to query, match, and output power outage areas, automatically and accurately determining safe work boundaries based on specific construction task characteristics. The power outage area classification table, built based on the mapping relationship between historical tasks and areas, ensures the standardization and consistency of area division. Finally, based on power outage areas, energized areas are divided into sets, completing the construction of the work environment risk space and providing a stable prediction space for subsequent dynamic behavior risk prediction.

[0108] Furthermore, by employing a work trajectory predictor with multiple branches, future movement trajectories for different key components are predicted in parallel. This approach comprehensively considers the specific tasks and individual personnel needs, making the prediction results more closely aligned with actual operational scenarios. Simultaneously, it provides a data foundation for the spatial relationship between work trajectory sequences and energized areas, as well as for risk detection.

[0109] Finally, by using operational experience data and insulation equipment wearing parameters, the error range of the initial predicted trajectory is adjusted. The preset error scale is dynamically adjusted using experience compensation coefficients and wearing compensation coefficients to improve the accuracy of risk assessment. A sequence set of operational trajectory ranges is generated, enhancing the realism and adaptability of the risk prediction model. The anomaly degree of each location is obtained by calculating the proportion of time the trajectory range is less than the safety threshold relative to the energized area. All anomalies are aggregated into an anomaly degree distribution. Combining spatial safety distance requirements with temporal probability statistics yields the anomaly degree distribution. Ultimately, the anomaly degree distribution serves as the identification result, driving early warning control and enabling timely safety interventions, thus improving the safety of power construction safety management.

[0110] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0111] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0112] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for identifying and controlling safe behaviors during power construction, characterized in that, The method includes: During the electrical construction of the distribution cabinet, images of the construction personnel are collected, and the wearing of insulating equipment is identified to obtain the distribution of insulating equipment wearing parameters. The distribution of insulating equipment wearing parameters includes wearing parameters at multiple key locations. When the distribution of the insulation equipment wearing parameters is qualified, the characteristics of the power construction task are obtained, and the power outage area and the live area are divided in the distribution cabinet; Based on the characteristics of the power construction task, the operation trajectory is predicted at multiple key locations to obtain a set of operation trajectory sequences; Based on the distribution of insulating equipment wearing parameters, the set of work trajectory sequences, and the energized area, combined with the work experience data of construction personnel, work trajectory compensation and anomaly prediction are performed to obtain the anomaly degree distribution, which is used as the identification result for early warning control.

2. The method for identifying and controlling safe behaviors during power construction according to claim 1, characterized in that, During electrical installation of the distribution cabinet, images of the construction personnel are collected, and their insulation equipment is identified to obtain the distribution of insulation equipment wearing parameters, including: During the electrical installation of the distribution cabinet, images of the construction workers are captured using the image acquisition equipment on the distribution cabinet. The personnel image is input into the insulating equipment wear recognition device, which identifies and outputs wearing parameters at multiple key locations to obtain the distribution of insulating equipment wearing parameters. The insulating equipment wear recognition device includes multiple attention recognition branches.

3. The method for identifying and controlling safe behaviors during power construction according to claim 2, characterized in that, The training steps for the insulating equipment wear identifier include: Based on historical testing data of insulating equipment wearing, a set of sample personnel images was collected, and the wearing parameters of multiple key locations in each sample personnel image were labeled to obtain multiple sets of sample wearing parameters, where each wearing parameter includes a wearing quality score. Based on a convolutional neural network, multiple attention recognition branches are constructed. The input of the multiple attention recognition branches is a person image, and the output is the wearing parameters of multiple key positions. Using the sample personnel image set and multiple sample wearing parameter sets, supervised training and testing are performed on multiple attention recognition branches respectively. After the test accuracy is qualified, the training of the insulating equipment wearing recognition device is completed.

4. The method for identifying and controlling safe behaviors during power construction according to claim 1, characterized in that, When the distribution of the insulation equipment wearing parameters is qualified, the characteristics of the power construction task are obtained, and the power outage area and the live area are divided in the distribution cabinet, including: Determine whether all wearing parameters within the distribution of wearing parameters of the insulating equipment meet the wearing parameter threshold. If yes, the distribution of wearing parameters of the insulating equipment is qualified; otherwise, it is unqualified and a prompt is given. When the distribution of the insulation equipment wearing parameters is qualified, the characteristics of the power construction task are obtained; Input the characteristics of the power construction task into the power outage area classification table, and output the power outage areas. The area outside the power outage zone within the distribution cabinet is demarcated to obtain the energized area.

5. The method for identifying and controlling safe behaviors during power construction according to claim 4, characterized in that, The power outage area classification table includes a mapping relationship between the sample power construction task feature set and the sample power outage area set. The power construction task features include the power construction scope.

6. The method for identifying and controlling safe behaviors during power construction according to claim 1, characterized in that, Based on the characteristics of the power construction task, the operation trajectory is predicted at multiple key locations to obtain a set of operation trajectory sequences, including: A work trajectory predictor is obtained, wherein the work trajectory predictor includes multiple work trajectory prediction branches corresponding to multiple key locations, wherein the input data of each work trajectory prediction branch is personnel characteristics and power construction task characteristics, and the output data is a sequence of work trajectories for key locations, and each work trajectory includes the spatial three-dimensional coordinates of the corresponding key location within the work time range. The characteristics of the construction personnel are obtained and combined with the characteristics of the power construction task, which are then input into the work trajectory predictor to predict the work trajectory sequence of multiple key locations, thus obtaining a set of work trajectory sequences of multiple key locations.

7. The method for identifying and controlling safe behaviors during power construction according to claim 1, characterized in that, Based on the distribution of insulating equipment wearing parameters, the set of work trajectory sequences, and the energized area, combined with the work experience data of construction personnel, work trajectory compensation and anomaly prediction are performed to obtain an anomaly degree distribution, which is used as the identification result for early warning control, including: Obtain work experience data from construction workers, including the percentage of abnormal work incidents. Based on the distribution of the insulating equipment wearing parameters and operational experience data, trajectory compensation is performed on the operational trajectory sequence set to obtain the operational trajectory range sequence set; Based on the set of operation trajectory range sequences and the energized area, anomaly prediction is performed to obtain the anomaly degree distribution; The anomaly distribution is used as the identification result for early warning control.

8. The method for identifying and controlling safe behaviors during power construction according to claim 7, characterized in that, Based on the distribution of insulating equipment wearing parameters and operational experience data, trajectory compensation is performed on the operational trajectory sequence set to obtain an operational trajectory range sequence set, including: Obtain the average percentage of abnormal operations in power construction, calculate the ratio of the percentage of abnormal operations in the operational experience data to the average percentage of abnormal operations, and obtain the experience compensation coefficient. Obtain the baseline wearing parameters, calculate the mean of the distribution of the wearing parameters of the insulating equipment, obtain the average wearing parameters of the construction personnel, and calculate the ratio of the baseline wearing parameters to the average wearing parameters to obtain the wearing compensation coefficient. A preset trajectory error scale is obtained, and the preset trajectory error scale is adjusted and calculated using the empirical compensation coefficient and the wearable compensation coefficient to obtain the adjusted trajectory error scale. The trajectory error scale is adjusted to compensate for the trajectory error of the operation trajectory sequence set to obtain the operation trajectory range sequence set. Each operation trajectory range includes the spatial three-dimensional coordinate interval of the corresponding key position within the operation time range.

9. The method for identifying and controlling safe behaviors during power construction according to claim 7, characterized in that, Based on the set of operation trajectory range sequences and the energized region, anomaly prediction is performed to obtain the anomaly degree distribution, including: Based on the set of operation trajectory range sequences and the energized area, the operation trajectory ranges that are less than the safe distance threshold from the energized area are extracted, and the proportion is calculated to obtain multiple anomalies. By aggregating multiple anomalies at multiple key locations, the anomaly distribution is obtained.

10. A power construction safety behavior identification and control device, characterized in that, The apparatus for implementing the power construction safety behavior identification and control method according to any one of claims 1-9, the apparatus comprising: The parameter distribution acquisition module is used to collect images of construction personnel during the power construction of the distribution cabinet, identify the wearing of insulating equipment, and obtain the wearing parameter distribution of the insulating equipment. The wearing parameter distribution of the insulating equipment includes wearing parameters at multiple key locations. The area division module is used to obtain the characteristics of the power construction task when the distribution of the insulation equipment wearing parameters is qualified, and to divide the power outage area and the live area in the distribution cabinet. The work trajectory prediction module is used to predict the work trajectory at multiple key locations based on the characteristics of the power construction task, and obtain a set of work trajectory sequences. The early warning control module is used to perform work trajectory compensation and anomaly prediction based on the distribution of the insulating equipment wearing parameters, the work trajectory sequence set, and the energized area, combined with the work experience data of the construction personnel, to obtain the anomaly degree distribution as the identification result and perform early warning control.

Images