AV-based field management method, system, device, medium and program product

Abstract

The invention relates to an AV-based field management method, system and device, a medium and a program product, and belongs to the field of intelligent management. According to the technical scheme, the method comprises the steps of inputting a project safety standard file into a natural language processing model, and extracting a safety facility standard; establishing a field digital twin 3D model; extracting field safety facility information according to the field digital twin 3D model; and comparing the on-site safety facility information with the safety facility standard, and determining the on-site safety facility which does not meet the safety facility standard in the on-site digital twin 3D model. According to the invention, effective fusion of safety facilities of a real physical world and safety management of the digital twinning 3D model can be realized, a virtual field digital twinning 3D model is enhanced, the safety management capability of the field digital twinning 3D model is enhanced, the safety management efficiency is improved, and the intelligent level of safety management is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent management technology, and in particular to an AV-based on-site management method, system, equipment, medium and program product. Background Art

[0002] In the current on-site management of new energy (wind turbines, photovoltaics, etc.) projects, based on the Internet of Things technology, a large number of sensors are used to return detection signals through the communication network, and then the detection signals are identified and judged to complete safety detection and management.

[0003] The shortcoming of the current on-site management methods for new energy projects is the lack of effective integration between the safety facilities in the real physical world and the safety management in the digital twin 3D model. This results in a mismatch between the safety management based on the digital twin 3D model and the real physical world, affecting the efficiency of safety management and restricting the improvement of the intelligent level of safety management.

[0004] Augmented Virtuality (AV) is a computer vision technology that superimposes elements of the real world in a virtual environment, seamlessly integrating real-world information into the virtual world. Summary of the Invention

[0005] In order to solve the above problems, the inventors have made the present invention, which provides an AV-based site management method, system, device, medium and program product.

[0006] In a first aspect, the present invention provides an AV-based on-site management method, comprising:

[0007] Input the project safety standard document into the natural language processing model to extract the safety facility standards;

[0008] Build a digital twin 3D model of the site;

[0009] Extracting on-site safety facility information based on the on-site digital twin 3D model;

[0010] The on-site safety facility information is compared with the safety facility standards to determine on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards.

[0011] Specifically, the project safety standards document is input into the natural language processing model to extract safety facility standards, including:

[0012] Collect project safety standard documents that match on-site projects;

[0013] Obtaining input text according to the project safety standard document;

[0014] Inputting the input text into a natural language processing model to obtain a first output text;

[0015] Annotate the first output text according to on-site project requirements;

[0016] The annotated first output text is input into the natural language processing model to obtain a second output text, where the second output text includes safety facility standards.

[0017] Specifically, the project safety standards document is fed into the natural language processing model to extract safety facility standards, including:

[0018] Iterating and optimizing the natural language processing model to obtain a production model;

[0019] Through knowledge distillation, the production model is compressed into a lightweight version to obtain a terminal deployment model.

[0020] Specifically, the on-site digital twin 3D model is established, including:

[0021] Collect real-time spatial data, attribute data and status data of safety facility layer and non-safety facility layer respectively;

[0022] An on-site digital twin 3D model is established based on the real-time spatial data, attribute data and status data.

[0023] Specifically, based on the on-site digital twin 3D model, on-site safety facility information is extracted, including:

[0024] Extracting real-time spatial data, attribute data, and status data of the safety facility layer from the on-site digital twin 3D model;

[0025] On-site safety facility information is extracted based on the real-time spatial data, attribute data, and status data of the safety facility layer. The on-site safety facility information includes the type, quantity, spatial location, and status of the on-site safety facilities.

[0026] Specifically, comparing the on-site safety facility information with the safety facility standards to determine on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards includes:

[0027] According to the safety facility standard, a first list is formed, wherein the fields in the first list include the type, quantity, spatial location and status of the safety facility;

[0028] forming a second list based on the on-site safety facility information, wherein the fields in the second list include the type, quantity, spatial location, and status of the on-site safety facility;

[0029] Compare the first list and the second list to determine on-site safety facilities that do not meet the safety facility standards and mark them in the on-site digital twin 3D model.

[0030] In a second aspect, the present invention provides an AV-based site management system, comprising:

[0031] The safety standards extraction module is used to input the project safety standards document into the natural language processing model to extract the safety facility standards;

[0032] A site model building module is used to build a site digital twin 3D model; based on the site digital twin 3D model, extract site safety facility information;

[0033] The on-site safety management module is used to compare the on-site safety facility information with the safety facility standards and determine the on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards.

[0034] Based on the same inventive concept, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the aforementioned AV-based on-site management method when executing the computer program.

[0035] Based on the same inventive concept, the present invention further provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and when the computer executable instructions are executed, the aforementioned AV-based site management method is implemented.

[0036] Based on the same inventive concept, the present invention further provides a computer program product, comprising a computer program / instruction, which implements the aforementioned AV-based site management method when executed by a processor.

[0037] The beneficial effects of the above technical solution provided by the present invention include at least:

[0038] By comparing the on-site safety facility information with the safety facility standards, the on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards are identified, thereby achieving an effective integration of the safety facilities in the real physical world and the safety management of the digital twin 3D model, enhancing the virtual on-site digital twin 3D model, and enhancing the safety management capabilities of the on-site digital twin 3D model, which helps to improve the efficiency of safety management and enhance the level of intelligent safety management.

[0039] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0040] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0042] Figure 1 Flowchart of the AV-based on-site management method in an embodiment of the present invention;

[0043] Figure 2 Schematic diagram of an AV-based on-site management system according to an embodiment of the present invention;

[0044] Figure 3 Schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0046] In order to solve the problems existing in the prior art, embodiments of the present invention provide an AV-based site management method, system, device, medium and program product.

[0047] The embodiment of the present invention provides an AV-based on-site management method, the process of which is as follows: Figure 1 As shown, the following steps are included:

[0048] Step 1: Input the project safety standards document into the natural language processing model to extract safety facility standards.

[0049] Standard documents related to safety facility standards are often scattered across multiple documents, such as the "Work Standards," "Management Standards," and "Technical Standards" for new energy projects. Extracting these safety facility standards requires a significant amount of time and expertise from these documents. Natural language processing technology can improve the efficiency of extracting safety facility standards. Natural language processing models are text processing models developed based on natural language processing technology. They can be used to learn the semantic features of input text and output structured document data. Pre-trained models such as BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly Optimized BERT Approach), and GPT (Generative Pre-trained Transformer) can be used to improve model convergence efficiency. Alternatively, other natural language processing models based on the Transformer architecture can be used. Pre-trained models significantly improve the efficiency and effectiveness of natural language processing tasks by pre-learning language features on large-scale general-purpose corpora.

[0050] In some specific embodiments, the project safety standards document is input into a natural language processing model to extract safety facility standards, including:

[0051] (1) Collect project safety standard documents that match the on-site project. For example, if the on-site project is a photovoltaic project, select standard documents related to photovoltaic projects from the standard documents such as "Work Standards", "Management Standards", and "Technical Standards" for new energy projects. If the on-site project is a wind power project, select standard documents related to wind power projects.

[0052] (2) Obtain input text based on the project safety standards document. Before inputting the original text of the standard document into the natural language processing model, the unstructured original text needs to be converted into a format that the model can process while retaining key semantic information. This includes: noise filtering, text normalization, and long text segmentation.

[0053] Noise filtering: Remove redundant elements from documents (such as headers, footers, and table comments) while retaining the core paragraph-level content. For example, in a safety standards document, remove the device parameter table but retain the textual description of the parameter requirements.

[0054] Text normalization: Standardize traditional and simplified Chinese fonts, full-width and half-width symbols, and correct typos. For specialized terms (such as "emergency evacuation channel"), build a domain dictionary to assist in subsequent word segmentation.

[0055] Long Text Segmentation: Text is segmented according to the maximum input length of the selected natural language processing model, typically 512 tokens (a token is the smallest unit of text segmentation in natural language processing, which can be translated as a word). A sliding window strategy is used to process long paragraphs to ensure contextual coherence. For example, an 800-word fire safety regulation paragraph can be split into two sections, with the first 50 tokens of the second section retained as the overlapping area to avoid information fragmentation.

[0056] Special processing: In Chinese scenarios, use a pre-trained Chinese word segmenter to dynamically segment unregistered words (such as newly emerged technical terms) or add them to a custom dictionary. For a pre-trained Chinese word segmenter, refer to the WordPiece solution in BERT-wwm-ext (BERT Whole Word Masking with Extensive Training).

[0057] Through the above preprocessing, the input text is obtained.

[0058] (3) Inputting the input text into a natural language processing model to obtain a first output text.

[0059] Model initialization and domain adaptation. Load a pre-trained model and adjust its parameters and structure based on task requirements, including base model selection, domain incremental training, and task header customization. Regarding base model selection, for example, RoBERTa-large or RoBERTa-wwm-ext is chosen as the base model, leveraging its deep semantic representation capabilities trained on a large corpus, including various technical documents, technical specifications, and management regulations in the electric power and energy sector. In domain incremental training, the pre-trained model is further trained on the target domain corpus (such as a collection of safety standards documents). Through the Masked Language Model (MLM) task, the model learns domain-specific expressions. For example, for sentences like "live video coverage rate 100%," the model's understanding of numerical ranges and unit combinations is strengthened. In task header customization, the model output layer is replaced based on downstream tasks (such as entity recognition and relation extraction). For example, in entity recognition tasks, a Bidirectional LSTM-CRF (Bidirectional LSTM Conditional Random Field) layer is added to capture entity boundaries and transition constraints between types through a CRF decoder.

[0060] Semantic feature extraction is performed, and the model forward propagation is used to obtain a deep semantic representation of the text, including contextual encoding, multi-granularity representation fusion, and implicit semantic reasoning. In contextual encoding, the input text is passed through RoBERTa's multi-head self-attention mechanism to capture long-range dependencies. For example, when analyzing the sentence "For non-removable sealed equipment, manholes must be installed, and the number of manholes must not be less than two," the model uses cross-sentence attention weights to associate the "manhole" in "sealed equipment" with specific numerical parameters. In multi-granularity representation fusion, the output of Transformers at different levels (such as word level and sentence level) is combined to enhance the ability to resolve ambiguous text. For example, in "The pressure value should be ≥1.2MPa," the model uses context to determine that the "≥" in "pressure value should be ≥1.2MPa" refers to a lower pressure limit rather than a comparative relationship. In implicit semantic reasoning, the model's ability to model negation words (such as "prohibited" and "should not") and conditional sentences (such as "if...then...") is used to infer implicit requirements beyond explicit rules. For example, from "communication facilities have been put into use, alarms have been installed and debugged, and are in normal use", it can be inferred that "early warning channels remain unobstructed".

[0061] Target semantic extraction accurately locates key information from the semantic representation output by the model, including entity recognition, relationship binding, and cross-sentence association. Entity recognition includes explicit entity extraction and implicit entity inference. Explicit entity extraction directly identifies named entities in the text (e.g., "emergency lighting" and "fire shutter door"). Implicit entity inference uses context to complete omitted components. For example, from the statement "evacuation distance must comply with regulations," the standard numerical value corresponding to "evacuation distance" can be inferred. Relationship binding includes rule-guided extraction and dependency parsing. In rule-guided extraction, prompt templates (prompts) are used to guide the model in generating structured relationships. For example, given the input "The installation height requirement for [facility] is ______," the model fills in "≥2.4 meters from the ground." Dependency parsing combines the syntax tree to determine the modification relationships between entities. For example, in the statement "At least one fire extinguisher is required for every 50 square meters," "every 50 square meters" is identified as a quantity adverbial, modifying "one fire extinguisher." In cross-sentence association, contextual information retained by the sliding window is used to establish a cross-paragraph logical chain. For example, the "photovoltaic panel sprinkler system coverage radius" and "installation spacing" requirements scattered in the same chapter are associated to form a complete configuration plan.

[0062] The model output is structured, converting the extracted semantics into machine-readable structured data to produce the first output text. This process involves information aggregation, which includes synonym merging and parameter standardization. Synonymous merging maps "firefighting equipment" and "fire extinguisher" into a unified entity. Parameter standardization converts "3 meters" into numeric data and adds a unit label.

[0063] (4) Annotate the first output text according to the requirements of the on-site project. By annotating the first output text, the accuracy of the model semantic extraction is improved. For example, accuracy is ensured through rule verification and annotation. In rule verification, domain specification matching and logical consistency detection are included. In domain specification matching, the standard file is compared to check whether the parameters meet the requirements, and the data that does not meet the requirements is annotated. In logical consistency detection, the structured semantic text output by the model is checked for logical conflicts or semantic redundancy, such as verifying whether "one fire extinguisher per 50 square meters" conflicts with "maximum protection distance 20 meters", and annotating the conflicting data. Through data annotation, during model training, the corresponding parameter weights can be adjusted to correct the output results and improve accuracy.

[0064] (5) Inputting the annotated first output text into the natural language processing model to obtain a second output text, wherein the second output text includes safety facility standards.

[0065] Furthermore, in some specific embodiments, the project safety standards document is input into a natural language processing model to extract safety facility standards, further comprising:

[0066] (a) Iterate and optimize the natural language processing model to obtain a production model. The production model is a model with better performance and higher accuracy formed after multiple iterations, optimization, and fine-tuning of the natural language processing model. It can be used in a formal production environment and is called a production model. Iterate and optimize the natural language processing model, for example, conduct error sample mining, count high-frequency mislabeled entities in manual review (such as misjudging "fire door" as "fire curtain"), and enhance training data in a targeted manner; conduct active model learning, and give priority to adding results with low confidence (such as "pressure value: 0.8~1.5MPa" is marked as uncertain) to the labeling queue. Through iteration and optimization, the model performance can be continuously improved and adapted to actual application needs.

[0067] (b) Through knowledge distillation, the production model is compressed into a lightweight version to obtain a terminal deployment model. Knowledge distillation is a model compression technology used to transfer the knowledge of a complex large model (called a teacher model) to a simpler model (called a student model). In traditional supervised learning, the model is trained only based on real labels; in knowledge distillation, the student model attempts to fit the output probability distribution of the teacher model under the same input. This mechanism enables the student model to capture more patterns and nuances, thereby achieving higher accuracy or better generalization ability. In the present application, the production model is the teacher model in knowledge distillation, and the terminal deployment model is the corresponding student model. RoBERTa-large is compressed into a lightweight version through knowledge distillation, which is convenient for meeting the real-time reasoning requirements of edge devices and facilitating distributed deployment. Furthermore, the distributed terminal deployment model is dynamically batched, its concurrent requests are merged and processed, and parallel computing devices such as GPUs are used to accelerate semantic extraction.

[0068] Furthermore, in some specific embodiments, the project safety standards document is input into a natural language processing model to extract safety facility standards, further comprising:

[0069] Continuously monitor the output text of natural language processing models to detect performance degradation and adapt to concept drift.

[0070] Perform performance degradation detection and retrain the model when degradation is detected. For example, regularly evaluate the model's F1 value on new documents and trigger retraining when a decrease is detected. The F1 value, also known as the F1-score, is the harmonic mean of precision and recall, used to comprehensively evaluate the performance of a binary classification model. The F1 value ranges from 0 to 1, with values ​​closer to 1 indicating better model predictions.

[0071] Concept drift refers to the phenomenon in which the statistical properties of a target variable change in unpredictable ways over time. Over time, the model's predictive accuracy decreases. To overcome the adverse effects of concept drift, concept drift adaptation is performed. This involves incrementally training the model to incorporate new semantic concepts when industry standards are updated (such as the addition of a new "IoT firefighting equipment" category). These new semantic concepts adjust the model's parameter weights, adapting the model to the new data and maintaining high predictive accuracy.

[0072] Step 2: Establish an on-site digital twin 3D model; extract on-site safety facility information based on the on-site digital twin 3D model.

[0073] In some specific embodiments, establishing a site digital twin 3D model includes:

[0074] (1) Collect real-time spatial data, attribute data, and status data of the safety facility layer and the non-safety facility layer respectively. Through sensors or computer vision recognition, collect real-time spatial data, attribute data, and status data of the on-site safety implementation layer. For example, through the GPS sensor arranged on the safety fence, collect the real-time spatial data of the safety fence. According to the correspondence between the GPS sensor number and the safety fence, collect the name of the safety facility corresponding to the GPS signal sent by the GPS sensor, that is, determine the corresponding attribute data. Through computer vision recognition, obtain the status of the safety fence at the spatial position. Its status is such as "normal" or "abnormal". The status of the safety fence such as deformation and damage can be classified as abnormal. Compare the real-time safety fence image with the normal safety fence image, and identify the status of the safety fence through visual detection algorithms such as YOLO.

[0075] Similarly, using the above method, real-time spatial data, attribute data, and status data are collected for the non-safety facility layer. The non-safety facility layer is a collection of other objects in the field that do not include safety facilities. Some objects in this collection are relevant to the evaluation of safety facilities, such as their distance from safety facilities and the number of safety facilities of a certain type within their area.

[0076] By establishing a layered on-site digital twin 3D model, in subsequent steps, it is only necessary to find on-site safety facilities that do not meet the safety facility standards in the safety facility layer, without having to search and detect the entire digital twin 3D model. Therefore, faster search and detection can be achieved, saving computing resources and time.

[0077] (2) Establishing an on-site digital twin 3D model based on the real-time spatial data, attribute data, and status data.

[0078] Specifically, based on real-time spatial data, basic physical model parameters such as the object's partial shape, size, and position are determined. Attribute data further determines the object's complete shape, name, and type. State data is used to determine the object's specific condition. Furthermore, based on real-time spatial data, attribute data, and state data, different types of objects in the on-site digital twin 3D model can be displayed with different colors or color borders for easier observation and management.

[0079] In some specific embodiments, extracting on-site safety facility information based on the on-site digital twin 3D model includes:

[0080] Extracting real-time spatial data, attribute data, and status data of the safety facility layer from the on-site digital twin 3D model;

[0081] On-site safety facility information is extracted based on the real-time spatial data, attribute data, and status data of the safety facility layer. The on-site safety facility information includes the type, quantity, spatial location, and status of the on-site safety facilities.

[0082] Since the real-time spatial data, attribute data and status data of the safety facility layer have been collected when establishing the on-site digital twin 3D model, the type, quantity, spatial location and status of the safety facilities in the digital twin 3D model can be determined through the safety facility layer data in the on-site digital twin 3D model.

[0083] Step 3: Compare the on-site safety facility information with the safety facility standards to determine the on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards.

[0084] In some specific embodiments, comparing the on-site safety facility information with the safety facility standard to determine on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standard includes:

[0085] (1) According to the safety facility standard, a first list is formed, wherein the fields in the first list include the safety facility type, quantity, spatial location and status.

[0086] For example, the first list is

[0087] Line number Type of safety facilities quantity Spatial location state 1001 ladder ≥2 Area 1, Area 2 normal 1002 railing ≥20 Area 1, Area 3 normal 1003 Hole cover ≥Number of holes Hole area normal

[0088] (2) Based on the on-site safety facility information, a second list is formed, wherein the fields in the second list include the type, quantity, spatial location, and status of the on-site safety facility.

[0089] An example of the second list is

[0090] Line number Type of safety facilities quantity Spatial location state 3d01 ladder ≥2 Area 1, Area 2 normal 3d02 railing ≥20 Area 1, Area 4 normal

[0091] (3) Compare the first list and the second list to determine the on-site safety facilities that do not meet the safety facility standards and mark them in the on-site digital twin 3D model.

[0092] Specifically, according to the "safety facility type" of each row in the first list, find the corresponding row in the second list, and then calculate the union of the quantity, spatial position and status in the corresponding rows of the first list and the second list respectively. If the union is the corresponding content in the first list, it meets the safety facility standards; otherwise, it is judged as not meeting the safety facility standards.

[0093] For example, based on the entry "ladder" in row 1001 of the first list, search for "safety facility type" in the second list and find its corresponding row, row 3d01 of the second list. The union of the "quantity" in row 1001 of the first list and row 3d01 of the second list is calculated. The union is "≥2," which corresponds to the "quantity" in row 1001 of the first list. Therefore, the "quantity" of "ladder" meets the safety facility standard. The "spatial location" and "status" are then determined. The union of the "spatial location" and "status" are "Area 1, Area 2" and "Normal," respectively, which correspond to the entries in row 1001 of the first list. Therefore, the "spatial location" and "status" of "ladder" meet the safety facility standard. Therefore, the on-site safety facility "ladder" meets the safety facility standard.

[0094] For example, based on the "railing" in row 1002 of the first list, search for the "safety facility type" in the second list and find its corresponding row in row 3d02 of the second list. The union of the "quantity" in row 1002 of the first list and row 3d02 of the second list is calculated. The union is "≥20," which is the "quantity" in row 1002 of the first list. Therefore, the "quantity" of "railings" meets the safety facility standards. Continue to determine the "spatial location" and "status." The union of the "spatial location" is "Area 1, Area 3, Area 4," which does not correspond to the content in row 1002 of the first list. Therefore, the "spatial location" of "railings" does not meet the safety facility standards. The union of the "status" is "normal," which is the corresponding content in row 1002 of the first list. Therefore, the "status" of "railings" meets the safety facility standards. Therefore, the on-site safety facility "railings" does not meet the safety facility standards.

[0095] On-site safety facilities that do not meet the safety facility standards will be marked in the on-site digital twin 3D model, for example, on-site safety facilities that do not meet the safety facility standards in the on-site digital twin 3D model will be marked with red boxes.

[0096] Furthermore, the on-site safety facility information is compared with the safety facility standards, and areas that do not meet the safety facility standards are displayed in the on-site digital twin 3D model.

[0097] If, by comparing the first list and the second list, the second list does not have a corresponding row of the first list, then based on the "spatial position" of the row in the first list, the area in the on-site digital twin 3D model that does not meet the safety facility standards is determined and marked with a red box. For example, according to the "hole cover" in row 1003 of the first list, there is no corresponding row in the second list, then based on the "spatial position" of row 1003 of the first list, the area in the on-site digital twin 3D model that does not meet the safety facility standards is determined to be a "hole area", which is highlighted in the on-site digital twin 3D model or marked with a red box. Through the above method, it is convenient for on-site safety management personnel to use the digital twin 3D model to identify safety facilities or related areas that do not meet the safety facility standards in a timely manner, thereby improving the level of intelligent safety management and improving the efficiency of safety management.

[0098] In the above method of this embodiment, the on-site safety facility information and the safety facility standards are compared to determine the on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards, thereby achieving an effective integration of the safety facilities in the real physical world and the safety management of the digital twin 3D model, enhancing the virtual on-site digital twin 3D model, and enhancing the safety management capabilities of the on-site digital twin 3D model, which helps to improve the efficiency of safety management and improve the level of intelligent safety management.

[0099] Those skilled in the art can change the above sequence without departing from the scope of protection of the present invention.

[0100] Another embodiment of the present invention provides an AV-based field management system, such as Figure 2 Shown, including:

[0101] The safety standards extraction module is used to input the project safety standards document into the natural language processing model to extract the safety facility standards;

[0102] A site model building module is used to build a site digital twin 3D model; based on the site digital twin 3D model, extract site safety facility information;

[0103] The on-site safety management module is used to compare the on-site safety facility information with the safety facility standards and determine the on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards.

[0104] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0105] The AV-based on-site management system in this embodiment realizes the effective integration of safety facilities in the real physical world and the safety management of the digital twin 3D model, enhances the virtual on-site digital twin 3D model, enhances the safety management capabilities of the on-site digital twin 3D model, helps to improve the efficiency of safety management, and improves the level of intelligent safety management.

[0106] Based on the same inventive concept, an embodiment of the present invention further provides an electronic device, the structure of which is as follows: Figure 3 As shown, it includes: a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the above-mentioned AV-based on-site management method is implemented.

[0107] Based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and when the computer executable instructions are executed by a processor, the aforementioned AV-based site management method is implemented.

[0108] Based on the same inventive concept, the present invention further provides a computer program product, comprising a computer program / instruction, which implements the aforementioned AV-based site management method when executed by a processor.

[0109] Any modifications, supplements and equivalent replacements made within the scope of the principles of the present invention shall still fall within the scope of the patent of the present invention. The "first" and "second" mentioned above only represent the distinction between different features.

Claims

1. An AV-based on-site management method, characterized in that: include: Input the project safety standard document into the natural language processing model to extract the safety facility standards; Build a digital twin 3D model of the site; Extracting on-site safety facility information based on the on-site digital twin 3D model; The on-site safety facility information is compared with the safety facility standards to determine on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards.

2. The method according to claim 1, wherein Input the project safety standards document into the natural language processing model to extract safety facility standards, including: Collect project safety standard documents that match on-site projects; Obtaining input text according to the project safety standard document; Inputting the input text into a natural language processing model to obtain a first output text; Annotate the first output text according to on-site project requirements; The annotated first output text is input into the natural language processing model to obtain a second output text, where the second output text includes safety facility standards.

3. The method according to claim 1, wherein Input the project safety standards document into the natural language processing model to extract safety facility standards, including: Iterating and optimizing the natural language processing model to obtain a production model; Through knowledge distillation, the production model is compressed into a lightweight version to obtain a terminal deployment model.

4. The method according to claim 1, wherein Build a 3D digital twin model of the site, including: Collect real-time spatial data, attribute data and status data of safety facility layer and non-safety facility layer respectively; An on-site digital twin 3D model is established based on the real-time spatial data, attribute data and status data.

5. The method according to claim 1, wherein Based on the on-site digital twin 3D model, extract on-site safety facility information, including: Extracting real-time spatial data, attribute data, and status data of the safety facility layer from the on-site digital twin 3D model; On-site safety facility information is extracted based on the real-time spatial data, attribute data, and status data of the safety facility layer. The on-site safety facility information includes the type, quantity, spatial location, and status of the on-site safety facilities.

6. The method according to claim 1, wherein Comparing the on-site safety facility information with the safety facility standards to determine on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards includes: According to the safety facility standard, a first list is formed, wherein the fields in the first list include the type, quantity, spatial location and status of the safety facility; forming a second list based on the on-site safety facility information, wherein the fields in the second list include the type, quantity, spatial location, and status of the on-site safety facility; Compare the first list and the second list to determine on-site safety facilities that do not meet the safety facility standards and mark them in the on-site digital twin 3D model.

7. An AV-based on-site management system, characterized in that: include: The safety standards extraction module is used to input the project safety standards document into the natural language processing model to extract the safety facility standards; A site model building module is used to build a site digital twin 3D model; based on the site digital twin 3D model, extract site safety facility information; The on-site safety management module is used to compare the on-site safety facility information with the safety facility standards and determine the on-site safety facilities in the on-site digital twin 3D model that do not meet the safety facility standards.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the AV-based site management method according to any one of claims 1 to 6 is implemented.

9. A computer storage medium, characterized in that The computer storage medium stores computer executable instructions, which, when executed, implement the AV-based on-site management method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the AV-based site management method according to any one of claims 1 to 6 is implemented.

Images