Power construction safety management method, device, equipment and medium

Abstract

The invention discloses a power construction safety management method and device, equipment and a medium, and belongs to the technical field of data processing, and the method comprises the steps: recognizing current construction site image data through a visual language model, outputting a dynamic description text and a scene text, comparing the dynamic description text and the scene text with a pre-obtained violation operation description text, and judging whether violation operation exists or not. Since the training and recognition process does not need to consider accident causes or accident probability, and only recognition similar to talking by looking at the image is carried out, the computing power requirement is low, the recognition efficiency and accuracy are high, the image information is converted into different texts for respective comparison, the complexity is reduced, the information transmission loss is reduced, and the user experience is improved. The comparison difficulty and the computing power requirement are further reduced, risk identification in safety management is focused on the aspect of violation operation, and the safety risk identification effect with half the effort is achieved.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method, device, equipment and medium for safety management of electric power construction. Background Art

[0002] Power construction is an important part of power system construction and maintenance, which covers all-round construction and maintenance work from power stations, substations to transmission lines. This process not only involves complex electrical installation and commissioning, but also includes multiple professional fields such as civil construction and equipment hoisting. In the field of power construction safety management, existing technical means have achieved certain results. For example, by installing sensors and monitoring equipment, changes in images or equipment parameters can be sensed in real time, and data can be transmitted to the monitoring center through wireless communication technology to achieve remote monitoring. At the same time, using the data analysis and processing system, the collected data can be quickly analyzed and judged to detect abnormal situations in a timely manner.

[0003] However, due to the complexity of the power construction site environment and the complexity of construction behaviors, it is difficult for existing technical means to make accurate judgments when identifying dangerous behaviors or safety risks. For example, the construction site has diverse topography and landforms, and the construction environment with different topography and landforms brings great difficulties to safety supervision. In addition, power construction involves multiple links such as high-altitude operations, power equipment operation, and construction material handling. Each link may have safety hazards. The identification of these safety hazards requires rich professional knowledge and practical experience. If the identification relies on neural network models for training, it requires huge training data and extremely high computing power requirements. Existing technical means are difficult to meet these conditions. Therefore, it is often difficult to fully cover and accurately identify in complex and changeable construction sites.

[0004] At the same time, since the intermediate processing of model training is a black box, based on the accident results, the model usually finds it difficult to distinguish whether the accident was caused by illegal operation or accident during training, and it is easy to capture wrong key information, resulting in a large amount of computing power being used in meaningless accident probability calculations, resulting in inefficient risk identification.

[0005] To this end, some existing technologies convert on-site images into text by image recognition and compare them with standard documents. However, in fact, due to the complex composition of the images, the difficulty of recognition has not been reduced, and the recognized text is also relatively complex. When directly comparing in this way, the accuracy is difficult to guarantee due to the accumulation of information loss in recognition and deviation in the comparison process. Therefore, the contradiction in the existing technology is that if a neural network model with black box properties is used to directly perform end-to-end recognition from the acquisition end to the output end, although there is almost no loss in information transmission, the computing power requirement is high and the efficiency is low; if segmented recognition is used, complex images and complex texts are prone to information transmission loss during the conversion and comparison process. Although the computing power requirement is not high, the accuracy is difficult to guarantee. Therefore, in the face of complex and changeable power construction sites, how to avoid the uncontrollability caused by black boxes and identify potential risks with the highest possible efficiency and accuracy while minimizing information transmission losses is one of the technical problems that have not yet been solved. Summary of the invention

[0006] In view of the problem that the recognition process in the prior art is uncontrollable or the information transmission loss is large, resulting in low accuracy and efficiency of risk identification, the present invention provides a method, device, equipment and medium for power construction safety management, which recognizes the current construction site image data through a visual language model, divides it into a dynamic description text and a scene description text, and compares the dynamic description text and the scene text with the pre-acquired illegal operation description text respectively, so as to determine whether there is an illegal operation. Since the training and recognition process of the present invention does not need to consider the cause of the accident or the probability of the accident, and only performs recognition similar to reading a picture, the computing power requirement is low, the recognition efficiency and accuracy are high, and the image information is converted into different texts for comparison to reduce complexity, reduce information transmission loss, further reduce the comparison difficulty and computing power requirements, and focus the risk identification in safety management on illegal operations, so as to achieve twice the result with half the effort in safety risk identification.

[0007] The following is the technical solution of the present invention.

[0008] The power construction safety management method includes the following steps: S1: According to the construction schedule, obtain the construction specification description text of different sub-items in the construction project at different construction schedules; S2: Based on the construction specification description text, antonym derivation is performed to obtain the illegal operation description text; S3: Acquire the current construction site image data in real time, use the trained visual language model to recognize the current construction site image data, and output dynamic description text and scene text; S4: Use the dynamic description text and the scene text to perform semantic recognition with the illegal operation description text respectively. If the recognition result meets the preset conditions, it is considered that the meaning is similar and a safety risk warning signal is output.

[0009] In the present invention, according to the construction schedule, the system will obtain the construction specification description text of each sub-item in the construction project at different construction stages to ensure the pertinence and timeliness of safety monitoring. By performing antonym derivation on the construction specification description text, the system can generate a description text of illegal operations to provide a comparison benchmark for subsequent violation detection. Then, the current construction site image data is identified through the visual language model, and dynamic description text and scene text are output to reduce complexity and reduce information transmission loss. This recognition, which is similar to picture-speaking, requires low computing power, high recognition efficiency and accuracy, and converts image information into text, which is then compared with the description text of illegal operations, further reducing the difficulty of comparison and computing power requirements, and achieving efficient and accurate safety risk identification.

[0010] Preferably, the S1: obtaining, according to the construction schedule, construction specification description texts of different sub-items in the construction project at different construction schedules, includes: Obtain normative documents including industry standards, project contracts, and technical manuals to form a normative text library; Divide the construction project into several sub-items and determine the construction period of each sub-item according to the construction period plan; According to the different construction schedules of each sub-item, the corresponding construction specification description text is matched from the specification text library, wherein the construction specification description text includes text content including standard processes, operating requirements and safety guidelines.

[0011] Preferably, the step S2: performing antonym derivation based on the construction specification description text to obtain the illegal operation description text includes: Use NLP technology to parse the collected construction specification description text, identify and extract key information, and store the parsed key information in a structured format; Based on the key information, define characteristics including sequence, restrictions, and conditions to describe compliance operations; Through logical reasoning or machine learning algorithms, the features describing the compliant operations are inverted to generate descriptive text for the illegal operations.

[0012] In the present invention, the redefined features can standardize the description text of the operation to reduce the difficulty of subsequent recognition and comparison.

[0013] Preferably, the S3: real-time acquisition of current construction site image data, recognition of current construction site image data using a trained visual language model, and output of dynamic description text and scene text includes: Based on the image sensors deployed at the construction site, the current construction site image data is acquired in real time, the received image data is pre-processed, and divided into multiple continuous frame segments with a preset time length; Based on the trained visual language model, each continuous frame segment is identified in turn, and dynamic description text including sequence, action, scene and scene text are output; When adjacent dynamic description texts are the same, increase the preset duration; otherwise, call back the preset duration to the initial value.

[0014] In the present invention, when adjacent dynamic description texts, that is, continuously output dynamic description texts, are identical, it means that there is no dynamic part in the picture or the dynamic part performs repeated operations, so the time interval between adjacent frames can be extended.

[0015] Preferably, the method of sequentially identifying each continuous frame segment based on the trained visual language model and outputting dynamic description text and scene text including sequence, action and scene includes: Compare each continuous frame segment to separate the dynamic area and the static area; Extract features from dynamic regions, obtain feature sequences in images, and model the extracted feature sequences to capture temporal dependencies and contextual information between consecutive frames. Generate dynamic description text including sequence and action information based on the results of sequence modeling; Feature extraction is performed on static areas to generate scene text.

[0016] Since construction specifications mostly exist in text form, image data of construction sites cannot be directly used to judge illegal operations. Therefore, the present invention uses a visual language model to identify and describe the construction site, converting the two data into the same type to facilitate comparative judgment.

[0017] Preferably, the training process of the visual language model includes: Collect image-text pair data related to power construction, clean, annotate and enhance the collected data to serve as training data; Build the M6 ​​model based on cross-modal information fusion, where: Use ResNet or EfficientNet as the visual encoder to extract spatial features and semantic information from images, and adjust the depth and width of the network according to computing resources and data size; Use the encoder part of Transformer as the text encoder to process the input description text or label, and set the number of layers, hidden units, and attention heads; Build a cross-modal attention mechanism or fusion layer to fuse the features extracted by the visual encoder and the text encoder to generate a cross-modal representation; Use the decoder part of Transformer to set the output layer according to the task requirements to generate description text; The training data is used to train and tune the M6 ​​model based on cross-modal information fusion to obtain the visual language model.

[0018] Preferably, the step S4: using the dynamic description text and the scene text to perform semantic recognition with the illegal operation description text respectively, if the recognition result meets the preset conditions, it is considered that the meanings are similar, and a safety risk warning signal is output, including: Based on the content of the scenario text, obtain the illegal operation description text corresponding to the construction progress, and calculate the cosine similarity based on the content of the dynamic description text and the restriction and condition information in the illegal operation description text. If the cosine similarity is higher than the threshold, it is considered that the meaning is similar, and a safety risk warning signal is output; If the cosine similarity is not higher than the threshold, the sequence information in the dynamic description text is extracted to obtain the sorting number of each action, and it is determined whether the sequence information in the illegal operation description text is met. If it is met, a security risk warning signal is output, otherwise it is considered that there is no illegal operation.

[0019] Preferably, when the security risk warning signal is output, a dynamic area or a static area where the risk is located is framed.

[0020] The present invention also provides an electric power construction safety management device, including field equipment, human-computer interaction equipment and processing equipment. When the field equipment, human-computer interaction equipment and processing equipment are in operation, the above-mentioned electric power construction safety management method is executed.

[0021] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned power construction safety management method when calling the computer program in the memory.

[0022] The present invention also provides a storage medium, in which computer executable instructions are stored. When the computer executable instructions are loaded and executed by a processor, the steps of the above-mentioned power construction safety management method are implemented.

[0023] The substantial effects of the present invention include: By combining the construction schedule with the construction specification description text, this method can accurately match the safety criteria of different construction stages, making safety monitoring more targeted and timely. This feature helps reduce safety risks caused by changes in the construction environment or different construction schedules.

[0024] NLP technology is used for antonym derivation to generate description text of illegal operations, providing the system with clear violation detection standards. Compared with traditional rule-based detection methods, this method is more flexible and adaptable, and can cope with complex and changing construction site environments.

[0025] The trained visual language model can be used to identify the image data of the construction site in real time and convert it into detailed dynamic description text and scene text. The dynamic description text and scene text are then semantically compared with the illegal operation description text to reduce complexity and information transmission loss, thus achieving efficient and accurate safety risk identification. Potential illegal operations can be quickly discovered and safety risk warning signals can be output in a timely manner, thereby effectively preventing the occurrence of safety accidents.

[0026] Therefore, the present invention can avoid using a large amount of computing resources for the identification or calculation of various accident details and probabilities, and complete the identification and early warning of illegal operations through simple identification and comparison, with higher efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution will be clearly and completely described below in combination with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0030] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0031] It should be understood that in the present invention, "plurality" refers to two or more than two. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "Contains A, B and C", "Contains A, B, C" means that A, B, and C are all included, "Contains A, B or C" means that one of A, B, and C is included, and "Contains A, B and / or C" means that any one, any two, or any three of A, B, and C are included.

[0032] The technical solution of the present invention is described in detail with specific embodiments below. The embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0033] Embodiment 1: Power construction safety management methods, such as Figure 1 As shown, the following steps are included: S1: According to the construction schedule, obtain the construction specification description text of different sub-items in the construction project at different construction progress.

[0034] Obtain normative documents including industry standards, project contracts, and technical manuals to form a normative text library; Divide the construction project into several sub-items and determine the construction period of each sub-item according to the construction period plan; According to the different construction schedules of each sub-item, the corresponding construction specification description text is matched from the specification text library, wherein the construction specification description text includes text content including standard processes, operating requirements and safety guidelines.

[0035] In this example, a comprehensive specification text library was first established. It not only integrates the latest industry standards, but also incorporates contract details and technical manuals for substation projects, ensuring the authority and pertinence of the specification content. Subsequently, the entire construction project was subdivided into several sub-items, such as transformer installation, high-voltage line laying, grounding system construction, etc., and the start and end time nodes were accurately set for each sub-item based on the overall project schedule.

[0036] This embodiment also carefully screens and matches the corresponding construction specification description texts from the specification text library according to the specific needs of each sub-item at different construction stages. These texts specify in detail the standard process for each stage, such as the hoisting steps and fixing requirements when installing the transformer, the cable specifications and laying methods when laying high-voltage lines, and the grounding resistance test and grounding material selection during the construction of the grounding system. At the same time, most importantly, the text also clearly lists the precautions and safety guidelines during the operation process, such as wearing insulating gloves and goggles, and observing electrical safety distances, to ensure that the construction process complies with the specifications and protect the personal safety of construction workers.

[0037] S2: Based on the construction specification description text, antonym derivation is performed to obtain the illegal operation description text.

[0038] Use NLP technology to parse the collected construction specification description text, identify and extract key information, and store the parsed key information in a structured format; Based on the key information, define characteristics including sequence, restrictions, and conditions to describe compliance operations; Through logical reasoning or machine learning algorithms, the features describing the compliant operations are inverted to generate descriptive text for the illegal operations.

[0039] In this embodiment, the redefined features can standardize the description text of the operation to reduce the difficulty of subsequent recognition and comparison. The order feature is used to describe the order between entities, the restriction feature is used to describe the rule restrictions of the entity (such as the need to wear a helmet, the distance must be more than 3 meters, etc.), and the conditional feature is used to describe the execution conditions of the entity (such as the transformer must be fixed for wiring operations).

[0040] Specifically in this embodiment, for example, there is a text about the construction specification of transformer installation in a substation: "Before installing the transformer, an appearance inspection must be carried out. After confirming that there is no damage, it can be fixed. After the fixing is completed, the wiring operation can be carried out. When wiring, it is necessary to ensure that the line connection is firm and well insulated." NLP technology will first pre-process this text, including steps such as removing punctuation marks, lowercase conversion, and word segmentation. Then, using the part-of-speech tagging function, each word in the text is marked with its corresponding part of speech, such as "transformer" is marked as a noun, "installation" is marked as a verb, etc.

[0041] Subsequently, key entities and the relationships between these entities are identified from the text to define features, such as the subject-predicate relationship between "transformer" and "installation", the sequence feature between "installation" and "appearance inspection", etc. These key information and entity relationships are then stored in a structured format, such as JSON or XML, to facilitate subsequent query, analysis, and processing. Through this parsing process, key information in the construction specification text is effectively extracted and sorted out, providing strong support for subsequent construction management and risk control.

[0042] Finally, the logical reasoning method was used to invert the features of the compliant operations and generate description texts of the illegal operations, such as "the transformer was not installed in the order of fixing first and then wiring", "the high-voltage line was not insulated", etc. This processing makes the identification of illegal operations more intuitive and accurate, providing strong support for subsequent safety risk judgment.

[0043] S3: Acquire the current construction site image data in real time, use the trained visual language model to recognize the current construction site image data, and output dynamic description text and scene text.

[0044] Based on the image sensors deployed at the construction site, the current construction site image data is acquired in real time, the received image data is pre-processed, and divided into multiple continuous frame segments with a preset time length; Compare each continuous frame segment to separate the dynamic area and the static area; Extract features from dynamic regions, obtain feature sequences in images, and model the extracted feature sequences to capture temporal dependencies and contextual information between consecutive frames. Generate dynamic description text including sequence and action information based on the results of sequence modeling; Extract features from static areas and generate scene text; when adjacent dynamic description texts are the same, increase the preset duration, otherwise call back the preset duration to the initial value.

[0045] Since construction specifications mostly exist in text form, image data of construction sites cannot be directly used to judge illegal operations. Therefore, the present invention uses a visual language model to identify and describe the construction site, converting the two data into the same type to facilitate comparative judgment.

[0046] For example, construction workers are operating a transformer to connect wires. After comparing consecutive frame segments, this part is identified as a dynamic area; while the surrounding substation equipment and environment remain stationary and are divided into a static area.

[0047] For dynamic areas, the system extracts the action feature sequences of construction workers, such as "construction workers hold wiring pliers and connect cables to transformer terminals", and models to capture the temporal dependencies and contextual information between these actions. At the same time, feature extraction is performed on static areas to generate scene text such as "inside the substation, the transformer is located in the center, surrounded by high-voltage switchgear and grounding devices".

[0048] In this embodiment, the training process of the visual language model includes: Collect image-text pair data related to power construction, clean, annotate and enhance the collected data to serve as training data; Build the M6 ​​model based on cross-modal information fusion, where: Use ResNet or EfficientNet as the visual encoder to extract spatial features and semantic information from images, and adjust the depth and width of the network according to computing resources and data size; Use the encoder part of Transformer as the text encoder to process the input description text or label, and set the number of layers, hidden units, and attention heads; Build a cross-modal attention mechanism or fusion layer to fuse the features extracted by the visual encoder and the text encoder to generate a cross-modal representation; Use the decoder part of Transformer to set the output layer according to the task requirements to generate description text; The training data is used to train and tune the M6 ​​model based on cross-modal information fusion to obtain the visual language model.

[0049] In this embodiment, the M6 ​​model is selected because it can process image and text data at the same time and realize cross-modal information fusion. In addition, the M6 ​​model supports multi-task learning and can perform multiple tasks such as image classification, target detection, and text generation at the same time to meet the diverse needs of power construction safety management. As a large Transformer model, M6 can process large-scale data sets and learn rich feature representations from them to improve the generalization ability and accuracy of the model. In addition, the structure of the M6 ​​model is flexible and can be customized and optimized according to specific needs, such as adjusting parameters such as the number of layers and the number of hidden units to meet the specific requirements of power construction scenarios.

[0050] Through the above model architecture and parameter settings, the M6 ​​model can make full use of the image and text data in the power construction scene, realize accurate image recognition and text generation, and provide strong technical support for the safety management of power construction.

[0051] S4: Use the dynamic description text and the scene text to perform semantic recognition with the illegal operation description text respectively. If the recognition result meets the preset conditions, it is considered that the meaning is similar and a safety risk warning signal is output.

[0052] Based on the content of the scenario text, obtain the illegal operation description text corresponding to the construction progress, and calculate the cosine similarity based on the content of the dynamic description text and the restriction and condition information in the illegal operation description text. If the cosine similarity is higher than the threshold, it is considered that the meaning is similar, and a safety risk warning signal is output; If the cosine similarity is not higher than the threshold, the sequence information in the dynamic description text is extracted to obtain the sorting number of each action, and it is determined whether the sequence information in the illegal operation description text is met. If it is met, a security risk warning signal is output, otherwise it is considered that there is no illegal operation.

[0053] In addition, when the security risk warning signal is output, the dynamic area or static area where the risk is located is selected.

[0054] For example, if the scene text is "transformer construction scene", the corresponding illegal operation description text is matched, and the dynamic description text is "construction workers wearing insulating gloves are performing wiring operations on fixed transformers", and the information in the dynamic description text is extracted.

[0055] The cosine similarity is calculated with the restriction and condition information in the description text of the illegal operation, such as the semantic recognition of "transformer", "wiring", "not wearing insulating gloves" in "not wearing insulating gloves when wiring the transformer". By calculating the cosine similarity between the two, when the similarity exceeds the preset threshold, such as 0.9, the system immediately determines that there is a safety risk and issues a warning signal.

[0056] If the cosine similarity does not meet the standard, the system will further analyze the sequence information in the site description, for example, first fix the transformer and then connect the wires. This sequence is compared with the "direct wiring without fixing the transformer" in the illegal operation. Since the sequence in the illegal operation is wiring before fixing, they do not match and no safety risk warning is output. If the sequence information in the site description is also wiring before fixing, a safety risk warning is also output. It should be noted that the sequence information can be extracted from the same text information, or it can be obtained by comparing the previous and next text information.

[0057] It can be seen that in this embodiment, according to the construction schedule, the system will obtain the construction specification description text of each sub-item in the construction project at different stages of the construction period to ensure the pertinence and timeliness of safety monitoring. By performing antonym derivation on the construction specification description text, the system can generate a description text of illegal operations to provide a comparison benchmark for subsequent violation detection. Next, the current construction site image data is recognized through the visual language model, and dynamic description text and scene text are output to reduce complexity and reduce information transmission losses. This recognition, which is similar to picture description, has low computing power requirements and high recognition efficiency and accuracy. The image information is converted into text and then compared with the illegal operation description text, which further reduces the comparison difficulty and computing power requirements, and realizes efficient and accurate safety risk identification.

[0058] Embodiment 2: This embodiment also provides a power construction safety management device, including field equipment, human-computer interaction equipment and processing equipment. When the field equipment, human-computer interaction equipment and processing equipment are running, the above-mentioned power construction safety management method is executed.

[0059] Embodiment 3: This embodiment also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned power construction safety management method when calling the computer program in the memory.

[0060] Embodiment 4: This embodiment also provides a storage medium, in which computer executable instructions are stored. When the computer executable instructions are loaded and executed by a processor, the steps of the above-mentioned power construction safety management method are implemented.

[0061] The substantial effects of this embodiment include: By combining the construction schedule with the construction specification description text, this method can accurately match the safety criteria of different construction stages, making safety monitoring more targeted and timely. This feature helps reduce safety risks caused by changes in the construction environment or different construction schedules.

[0062] NLP technology is used for antonym derivation to generate description text of illegal operations, providing the system with clear violation detection standards. Compared with traditional rule-based detection methods, this method is more flexible and adaptable, and can cope with complex and changing construction site environments.

[0063] The trained visual language model can be used to identify the image data of the construction site in real time and convert it into detailed dynamic description text and scene text. The dynamic description text and scene text are then semantically compared with the illegal operation description text to reduce complexity and information transmission loss, thus achieving efficient and accurate safety risk identification. Potential illegal operations can be quickly discovered and safety risk warning signals can be output in a timely manner, thereby effectively preventing the occurrence of safety accidents.

[0064] Therefore, this embodiment can avoid using a large amount of computing resources for the identification or calculation of various accident details and probabilities, and complete the identification and early warning of illegal operations through simple identification and comparison, with higher efficiency and accuracy.

[0065] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the specific device can be divided into different functional modules to complete all or part of the functions described above.

[0066] In the embodiments provided in the present application, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the embodiments of the structure described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another structure, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, structures or units, which can be electrical, mechanical or other forms.

[0067] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0068] In addition, each functional unit in the embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0069] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0070] The above contents are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for safety management of electric power construction, characterized in that: The following steps are involved: S1: According to the construction schedule, obtain the construction specification description text of different sub-items in the construction project at different construction schedules; S2: Based on the construction specification description text, antonym derivation is performed to obtain the illegal operation description text; S3: Acquire the current construction site image data in real time, use the trained visual language model to recognize the current construction site image data, and output dynamic description text and scene text; S4: Use the dynamic description text and the scene text to perform semantic recognition with the illegal operation description text respectively. If the recognition result meets the preset conditions, it is considered that the meaning is similar and a safety risk warning signal is output.

2. The power construction safety management method according to claim 1, characterized in that: S1: According to the construction schedule, obtain the construction specification description text of different sub-items in the construction project at different construction schedules, including: Obtain normative documents including industry standards, project contracts, and technical manuals to form a normative text library; Divide the construction project into several sub-items and determine the construction period of each sub-item according to the construction period plan; According to the different construction schedules of each sub-item, the corresponding construction specification description text is matched from the specification text library, wherein the construction specification description text includes text content including standard processes, operating requirements and safety guidelines.

3. The power construction safety management method according to claim 1, characterized in that: S2: Based on the construction specification description text, antonym derivation is performed to obtain the illegal operation description text, including: Use NLP technology to parse the collected construction specification description text, identify and extract key information, and store the parsed key information in a structured format; Based on the key information, define characteristics including sequence, restrictions, and conditions to describe compliance operations; Through logical reasoning or machine learning algorithms, the features describing the compliant operations are inverted to generate descriptive text for the illegal operations.

4. The power construction safety management method according to claim 1, characterized in that: S3: real-time acquisition of current construction site image data, recognition of current construction site image data using a trained visual language model, and output of dynamic description text and scene text, including: Based on the image sensors deployed at the construction site, the current construction site image data is acquired in real time, the received image data is pre-processed, and divided into multiple continuous frame segments with a preset time length; Based on the trained visual language model, each continuous frame segment is identified in turn, and dynamic description text including sequence, action, scene and scene text are output; When adjacent dynamic description texts are the same, increase the preset duration; otherwise, call back the preset duration to the initial value.

5. The power construction safety management method according to claim 4, characterized in that: Based on the trained visual language model, each continuous frame segment is identified in turn, and dynamic description text including sequence, action, scene and scene text are output, including: Compare each continuous frame segment to separate the dynamic area and the static area; Extract features from dynamic regions, obtain feature sequences in images, and model the extracted feature sequences to capture temporal dependencies and contextual information between consecutive frames. Generate dynamic description text including sequence and action information based on the results of sequence modeling; Feature extraction is performed on static areas to generate scene text.

6. The power construction safety management method according to claim 1, 4 or 5, characterized in that: The training process of the visual language model includes: Collect image-text pair data related to power construction, clean, annotate and enhance the collected data to serve as training data; Build the M6 ​​model based on cross-modal information fusion, where: Use ResNet or EfficientNet as the visual encoder to extract spatial features and semantic information from images, and adjust the depth and width of the network according to computing resources and data size; Use the encoder part of Transformer as the text encoder to process the input description text or label, and set the number of layers, hidden units, and attention heads; Build a cross-modal attention mechanism or fusion layer to fuse the features extracted by the visual encoder and the text encoder to generate a cross-modal representation; Use the decoder part of Transformer to set the output layer according to the task requirements to generate description text; The training data is used to train and tune the M6 ​​model based on cross-modal information fusion to obtain the visual language model.

7. The power construction safety management method according to claim 5, characterized in that: S4: using the dynamic description text and the scene text to perform semantic recognition with the illegal operation description text respectively, if the recognition result meets the preset conditions, it is considered that the meaning is similar, and a safety risk warning signal is output, including: Based on the content of the scenario text, obtain the illegal operation description text corresponding to the construction progress, and calculate the cosine similarity based on the content of the dynamic description text and the restriction and condition information in the illegal operation description text. If the cosine similarity is higher than the threshold, it is considered that the meaning is similar, and a safety risk warning signal is output; If the cosine similarity is not higher than the threshold, the sequence information in the dynamic description text is extracted to obtain the sorting number of each action, and it is determined whether the sequence information in the illegal operation description text is met. If it is met, a security risk warning signal is output, otherwise it is considered that there is no illegal operation.

8. The power construction safety management method according to claim 7, characterized in that: When the security risk warning signal is output, a dynamic area or a static area where the risk is located is framed.

9. An electric power construction safety management device, comprising field equipment, human-machine interaction equipment and processing equipment, characterized in that: When the field equipment, human-computer interaction equipment and processing equipment are in operation, the power construction safety management method as described in any one of claims 1 to 8 is executed.

10. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the power construction safety management method as claimed in any one of claims 1 to 8 when calling the computer program in the memory.

11. A storage medium, characterized in that: The storage medium stores computer executable instructions, and when the computer executable instructions are loaded and executed by the processor, the steps of the power construction safety management method as described in any one of claims 1 to 8 are implemented.

Images