Intelligent analysis method and system for DWG drawing of power system

Abstract

The invention discloses an intelligent analysis method for a DWG drawing of an electric power system. The method comprises the steps of obtaining a to-be-analyzed target DWG drawing of the electric power system and preprocessing the to-be-analyzed target DWG drawing; constructing an equipment symbol detection model based on deep learning and a graph neural network; performing symbol and parameter feature extraction on the obtained data information by adopting the constructed model, and performing parameter and symbol association; processing the obtained data information based on a graph attention network so as to reconstruct a topological relation; and carrying out integration, packaging and format conversion on the obtained data information to finish the analysis of the DWG drawing of the target power system. The invention also discloses a system for realizing the intelligent analysis method for the DWG drawing of the power system. According to the method, intelligent analysis of the DWG drawing of the power system is realized, the reliability is higher, and the accuracy is better.

Description

Technical Field

[0001] This invention belongs to the field of electrical automation, and specifically relates to an intelligent parsing method and system for power system DWG drawings. Background Technology

[0002] With economic and technological development and the improvement of people's living standards, electricity has become an indispensable secondary energy source in people's production and daily life, bringing endless convenience. Therefore, ensuring a stable and reliable supply of electricity has become one of the most important tasks of the power system.

[0003] In engineering fields such as primary electrical design of substations, designers first create two-dimensional (DWG) format engineering drawings. These drawings contain a wealth of graphic and non-graphical information, such as standardized symbols for circuit breakers, transformers, and disconnectors; textual annotations of equipment model specifications and voltage levels; precise dimensions and location information; and topological structures defining the electrical connections between equipment. Then, engineers need to interpret this information from the drawings.

[0004] Currently, the interpretation of drawing information still relies on manual methods. However, manual interpretation is not only time-consuming and labor-intensive, but also suffers from poor reliability and accuracy. In recent years, with the development of deep learning technology, researchers have proposed drawing information interpretation schemes based on neural network models, such as the scheme with application number CN202510595157.1. However, these schemes have limited accuracy in interpreting symbols that require understanding the global context layout for accurate classification; moreover, these schemes typically treat symbol recognition, parameter extraction, and topological analysis as independent and fragmented steps, lacking an end-to-end holistic solution, leading to information association errors and failing to form a complete and accurate digital model. Summary of the Invention

[0005] One of the objectives of this invention is to provide a highly reliable and accurate intelligent parsing method for power system DWG drawings.

[0006] The second objective of this invention is to provide a system for implementing an intelligent parsing method for power system DWG drawings.

[0007] The intelligent parsing method for power system DWG drawings provided by this invention includes the following steps:

[0008] S1. Obtain the DWG drawing of the target power system to be parsed;

[0009] S2. Preprocess the DWG drawings obtained in step S1;

[0010] S3. Construct a device symbol detection model based on deep learning and graph neural networks;

[0011] S4. Using the model constructed in step S3, extract symbolic and parameter features from the data information obtained in step S2, and associate the parameters and symbols.

[0012] S5. Based on the graph attention network, process the data information obtained in step S4 to reconstruct the topological relationships;

[0013] S6. Integrate, encapsulate, and convert the data information obtained in step S5 to complete the parsing of the target power system DWG drawing.

[0014] Step S2, which involves preprocessing the DWG drawing obtained in step S1, specifically includes the following steps:

[0015] Load DWG drawings;

[0016] Traverse all entities in the model space and analyze the geometric primitive features and non-geometric primitive features of the entity model;

[0017] For a defined block in a spatial entity, the original primitives included in the defined block are parsed to determine the geometric primitive features and non-geometric primitive features in the defined block;

[0018] All the entity information obtained from the parsing is stored in a structured data container to obtain a structured vector dataset; where each record represents a primitive and each record includes all geometric attribute information and non-geometric data information.

[0019] Based on the obtained structured vector dataset, generate structured vector data for image geometric calculations and topological relationship construction;

[0020] A raster image is generated based on the obtained structured vector data.

[0021] In the process of analyzing geometric primitive features: for lines, extract the 3D coordinates of the starting point and the ending point; for circles, extract the 3D coordinates of the center and the radius; for arcs, extract the 3D coordinates of the center, the radius, the starting angle, and the ending angle; for polylines, extract the 3D coordinates of all vertices and the closing symbol; for blocks, extract the name of the corresponding block, the position of the symbol in the drawing, the scaling ratio, and the rotation angle; for text, extract the string, the 3D coordinates of the insertion point in the text, the text height, and the layer it belongs to.

[0022] In the parsing of non-geometric primitive features: for layers, extract the name of the layer to which the entity belongs; for colors, extract the color index or true color value of the entity; for line types, extract the line type name.

[0023] The following steps are used to generate structured vector data for image geometric calculations and topological relation construction:

[0024] Data cleaning and filtering: Based on the extracted layer and entity type information, the vector data is filtered to remove entities on the Defpoints layer or irrelevant annotation layers;

[0025] Geometric data standardization: unify all coordinates to a unified coordinate system, normalize the rotation angle of the text, and calculate the bounding box of each device symbol;

[0026] Topology construction preparation: Based on the block definition of device symbols, the positions of corresponding electrical connection points or ports are predefined or learned, and the wire primitives are preprocessed; the preprocessing includes discretizing continuous LWPOLYLINE into a sequence of line segments;

[0027] Data output: Output structured vector data;

[0028] The following steps are used to generate a raster image:

[0029] Render View: Select the range of the model space for rendering the DWG file or specify the layout, calculate the bounding boxes of all entities to determine the area that needs to be rendered;

[0030] Set rendering parameters: Set the resolution to the set value; calculate the pixel size of the image based on the rendering area and the set resolution;

[0031] Perform rendering: Using the rendering engine that is compatible with the parsing library, generate the corresponding raster image based on the set rendering parameters;

[0032] Metadata association: Records rendering parameters, outputs raster images and their corresponding mappings to vector coordinates.

[0033] Step S3, which describes building a device symbol detection model based on deep learning and graph neural networks, specifically includes the following steps:

[0034] Constructing a CNN branch based on ResNet: The top-level global average pooling layer and fully connected layer in the ResNet convolutional neural network are removed, and the remaining ResNet network is used as a CNN branch; the CNN branch is used to extract local detail features of device symbols;

[0035] The Transformer architecture is adopted, and Transformer branches are constructed. Transformer branches are used to capture the global layout of the input image and the long-distance dependencies between symbols.

[0036] The outputs of the CNN branch and the Transformer branch are added element-wise, and then passed through a... The convolutional layers are processed to obtain the output of the device symbol detection model.

[0037] Step S4, using the model constructed in step S3, extracts symbolic and parameter features from the data information obtained in step S2 and associates the parameters and symbols, specifically including the following steps:

[0038] The raster image obtained in step S2 is input into the model constructed in step S3 to obtain the device symbol detection result output by the model.

[0039] For each detected device symbol, calculate the bounding box, category label, and confidence score; for each detected text, perform text recognition and output the text content and bounding box.

[0040] Unify the bounding boxes of device symbols and text bounding boxes to the set coordinate system and perform associated calculations;

[0041] For each device symbol, based on the association calculation results, select the text corresponding to the text box with the highest association score as the main parameter of that device symbol; at the same time, it is set that a text can only be assigned to one device symbol.

[0042] Step S5, based on the graph attention network, processes the data information obtained in step S4 to reconstruct the topological relationships, specifically including the following steps:

[0043] Convert the geometric information of the DWG drawing into a graph data structure;

[0044] Based on geometric information, all possible connection relationships are obtained by filtering.

[0045] All possible connections are processed using a graph attention network to obtain the connection probabilities of different importance between different nodes in the graph data structure.

[0046] Based on the obtained connection probabilities, a connection list is constructed; the connection list includes all devices in the DWG drawing, as well as the electrical connections between all devices.

[0047] Step S6 involves integrating, encapsulating, and converting the data information obtained in step S5 to complete the parsing of the target power system DWG drawing. Specifically, this includes the following steps:

[0048] All the obtained data is integrated into a defined data structure; all the data includes symbol identification results, parameter extraction results, symbol connection relationships, and symbol-parameter association relationships.

[0049] The obtained data results are mapped to a standard data format to ensure data universality;

[0050] Complete the analysis of the target power system DWG drawings.

[0051] This invention also provides a system for implementing the intelligent parsing method of the power system DWG drawing, comprising a drawing acquisition module, a drawing processing module, a model building module, a drawing association module, a topology reconstruction module, and a drawing parsing module; the drawing acquisition module, drawing processing module, model building module, drawing association module, topology reconstruction module, and drawing parsing module are connected in series; the drawing acquisition module is used to acquire the target power system DWG drawing to be parsed and upload the data information to the drawing processing module; the drawing processing module is used to preprocess the acquired DWG drawing according to the received data information and upload the data information to the model building module; the model building module is used to, based on the received data information, perform... Deep learning and graph neural networks are used to construct a device symbol detection model, and the data information is uploaded to the drawing association module. The drawing association module is used to extract symbol and parameter features from the received data information using the constructed model, associate parameters and symbols, and upload the data information to the topology reconstruction module. The topology reconstruction module is used to process the received data information based on a graph attention network to reconstruct the topology relationship, and upload the data information to the drawing parsing module. The drawing parsing module is used to integrate, encapsulate, and convert the format of the received data information to complete the parsing of the target power system DWG drawing.

[0052] The intelligent parsing method and system for power system DWG drawings provided by this invention, through processing of DWG drawings and based on deep learning networks, graph neural networks, graph attention networks and correlation analysis, not only achieves intelligent parsing of power system DWG drawings, but also has higher reliability and better accuracy. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0054] Figure 2 This is a schematic diagram of the functional modules of the system of the present invention. Detailed Implementation

[0055] like Figure 1 The diagram shown is a flowchart of the method of the present invention: The intelligent parsing method for power system DWG drawings disclosed in this invention includes the following steps:

[0056] S1. Obtain the DWG drawing of the target power system to be parsed;

[0057] S2. Preprocess the DWG drawing obtained in step S1; specifically including the following steps:

[0058] Loading DWG drawings; for example, the ezdxf development library can be used to load DWG files. First, the version of the DWG file is checked, and the parsing library automatically performs version adaptation to ensure that different versions of drawings can be read correctly.

[0059] Iterate through all entities in the model space and parse the geometric primitive features and non-geometric primitive features of the entity model; for example, use the parser of the ezdxf development library to traverse all entities in the model space.

[0060] For a defined block in a spatial entity, the original primitives included in the defined block are parsed to determine the geometric primitive features and non-geometric primitive features in the defined block;

[0061] All the parsed entity information is stored in a structured data container to obtain a structured vector dataset; each record represents a primitive, and each record includes all geometric attribute information and non-geometric data information; the data container can be a Python list dictionary, Pandas DataFrame, or a dedicated geospatial database, etc.

[0062] Based on the obtained structured vector dataset, generate structured vector data for image geometric calculations and topological relationship construction;

[0063] Based on the obtained structured vector data, generate high-quality, high-resolution raster images;

[0064] In practice, the ODA plugin is installed in the ezdxf development library. The ODA plugin is used in ezdxf to parse CAD files, mainly to handle newer or specific versions of DWG files that may not be supported by the built-in parser of ezdxf. The ODA plugin provides a library called Teigha, which can read and write DWG and DXF files. ezdxf converts the DWG file to DXF by calling the conversion tool of ODA, and then uses ezdxf to parse the DXF content. This can make full use of the ODA plugin of ezdxf to deeply parse DWG files and extract key data such as device symbols, geometric information, and text parameters, providing high-quality input data for subsequent hybrid deep learning models and topology inference.

[0065] In practice, during the analysis of geometric primitive features: for lines, extract the 3D coordinates of the starting point and the ending point; for circles, extract the 3D coordinates of the center and the radius; for arcs, extract the 3D coordinates of the center, the radius, the starting angle, and the ending angle; for polylines, extract the 3D coordinates of all vertices and the closing symbol; for blocks, extract the name of the corresponding block, the position of the symbol in the drawing, the scaling ratio, and the rotation angle; for text, extract the string, the 3D coordinates of the insertion point in the text, the text height, and the layer it belongs to.

[0066] In the parsing of non-geometric primitive features: for layers, extract the name of the layer to which the entity belongs for entity classification; for colors, extract the color index or true color value of the entity; for line types, extract the line type name.

[0067] The following steps are used to generate structured vector data for image geometric calculations and topological relation construction:

[0068] Data cleaning and filtering: Based on the extracted layer and entity type information, the vector data is filtered to remove entities on the Defpoints layer or irrelevant annotation layers;

[0069] Geometric data standardization: unify all coordinates to a unified coordinate system, normalize the rotation angle of the text, and calculate the bounding box of each device symbol;

[0070] Topology construction preparation: Based on the block definition of device symbols, the positions of corresponding electrical connection points or ports are predefined or learned, and the conductor primitives are preprocessed; the preprocessing includes discretizing continuous LWPOLYLINE into a sequence of line segments for subsequent calculation of spatial relationships with device ports;

[0071] Data output: Output structured vector data;

[0072] The following steps are used to generate a raster image:

[0073] Render View: Select the range of the model space for rendering the DWG file or specify the layout, calculate the bounding boxes of all entities to determine the area that needs to be rendered;

[0074] Set rendering parameters: Set the resolution to the set value, ensuring that small text and device symbols remain clearly visible in the image; calculate the pixel size of the image based on the rendering area and the set resolution; the preferred solution is to render with a fixed fine line width as a 3-channel RGB color image with a white background.

[0075] Perform rendering: Using the rendering engine that is compatible with the parsing library, generate the corresponding raster image based on the set rendering parameters, preferably a PNG or JPEG format raster image;

[0076] Metadata association: Records rendering parameters, outputs raster images and their corresponding mappings to vector coordinates;

[0077] S3. Construct a device symbol detection model based on deep learning and graph neural networks; specifically including the following steps:

[0078] Constructing a CNN branch based on ResNet: The top-level global average pooling layer and fully connected layer in the ResNet convolutional neural network are removed, and the remaining ResNet network is used as a CNN branch; the CNN branch is used to extract local detail features of device symbols;

[0079] The CNN branch's processing includes: the input image is processed by ResNet, which outputs a feature pyramid network, i.e., feature maps at different scales, used to detect device symbols of different sizes; the feature pyramid upsamples the output feature maps from top to bottom, and then connects the upsampled feature maps laterally with those from the shallow layers of the backbone network after channel alignment through 1x1 convolutions, outputting a set of multi-scale feature maps that fuse high-level semantic information and low-level positional information. This multi-scale feature map is the final output of the CNN branch.

[0080] The Transformer architecture is adopted, and Transformer branches are constructed. Transformer branches are used to capture the global layout of the input image and the long-distance dependencies between symbols.

[0081] The outputs of the CNN branch and the Transformer branch are added element-wise, and then passed through a... The convolutional layers are processed to obtain the output of the device symbol detection model;

[0082] S4. Using the model constructed in step S3, extract symbolic and parameter features from the data information obtained in step S2, and correlate the parameters and symbols; specifically including the following steps:

[0083] The raster image obtained in step S2 is input into the model constructed in step S3 to obtain the device symbol detection result output by the model.

[0084] For each detected device symbol, calculate the bounding box, category label, and confidence score; for each detected text, perform text recognition and output the text content and bounding box.

[0085] Unify the bounding boxes of device symbols and text bounding boxes to the set coordinate system and perform associated calculations;

[0086] For each device symbol, based on the association calculation results, select the text corresponding to the text box with the highest association score as the main parameter of that device symbol; at the same time, it is set that: a text can only be assigned to one device symbol;

[0087] In practice, the symbol bounding boxes output by the model and the text bounding boxes output by the OCR engine are unified to the same pixel coordinate system. Spatial proximity analysis is performed for each device symbol to initially filter out texts that may be related to it. For each device symbol, a dynamic search area is defined with its bounding box as the center. This area can be a circle with a fixed pixel radius or a rectangular area formed by expanding the symbol bounding box outward by a certain proportion. All text bounding boxes whose center points fall within this search area are identified, and a candidate text list is output. Semantic rule filtering uses prior knowledge from the engineering field to remove obviously irrelevant items from the candidate texts and prioritizes the remaining items. It checks whether the text content conforms to the common paradigm of device numbers. Texts that conform to the pattern have the highest priority. It identifies whether the text contains explicit technical parameter units or keywords, such as voltage in KV, current in A, etc. Device numbers are usually located above, below, to the left or right of the symbol and are very close. Parameter texts, such as voltage levels, may be slightly further away or point to the symbol through leads. If the OCR can provide feedback on the DWG layer where the text is located, it can prioritize associating texts on the same logical layer as the device symbol.

[0088] S5. Based on the graph attention network, process the data information obtained in step S4 to reconstruct the topological relationships; specifically including the following steps:

[0089] Convert the geometric information of the DWG drawing into a graph data structure;

[0090] Based on geometric information, all possible connection relationships are obtained by filtering.

[0091] All possible connections are processed using a graph attention network to obtain the connection probabilities of different importance between different nodes in the graph data structure.

[0092] Based on the obtained connection probabilities, a connection list is constructed; the connection list includes all devices in the DWG drawings, as well as the electrical connections between all devices; the purpose of this step is to provide the core data foundation for generating a computable BIM model, performing power flow calculations, or constructing a digital twin.

[0093] In practical implementation, during the training process of GNN, basic physical constraints of electrical engineering can be added as rules or part of the loss function. A circuit breaker usually has only two effective connection ports. If GNN predicts three high-probability connections for the same circuit breaker node, the system can introduce rules to retain only the two with the highest probabilities, or add a penalty term to the loss function to allow the model to learn this constraint during training. By using graph neural networks, the complex and error-prone geometric connection judgment problem is transformed into a more robust and accurate data-driven connection prediction problem, realizing deep intelligent analysis of the topological relationships of engineering drawings.

[0094] S6. Integrate, encapsulate, and convert the data information obtained in step S5 to complete the parsing of the target power system DWG drawing; specifically including the following steps:

[0095] All the obtained data is integrated into a defined data structure, which is a complete, self-describing JSON / XML file containing all the results of the drawing analysis; all the data includes symbol identification results, parameter extraction results, symbol connection relationships, and symbol-parameter association relationships;

[0096] The obtained data results are mapped to a standard data format to ensure data universality;

[0097] Complete the analysis of the target power system DWG drawings.

[0098] Furthermore, after completing the parsing of DEG drawings, the parsing results can be delivered to the platform to verify the effect: using the IFC SDK or a custom writer, the parsed data is generated into an .ifc file; the generated .ifc file is then provided directly to engineers, who can manually import it into the BIM software. Through the BIM platform's API, the structured data is directly pushed to the running BIM project, achieving true seamless automation; simultaneously, within the BIM platform, equipment models and connection networks with electrical attributes are automatically generated or updated based on the .ifc file. On this basis, engineers can perform clash detection, pipeline integration inspection, quantity surveying, and simulation analysis.

[0099] The present invention utilizes vector and raster dual-path parallel processing technology to perform in-depth analysis and preprocessing of DWG drawings, and uses a hybrid CNN-Transformer model to collaboratively complete the detection, classification and extraction of associated parameters of equipment symbols; therefore, the method of the present invention can better perform intelligent analysis of power system DWG drawings.

[0100] like Figure 2The diagram shows the functional modules of the system of the present invention: The system for implementing the intelligent parsing method of the power system DWG drawing disclosed in this invention includes a drawing acquisition module, a drawing processing module, a model building module, a drawing association module, a topology reconstruction module, and a drawing parsing module; the drawing acquisition module, drawing processing module, model building module, drawing association module, topology reconstruction module, and drawing parsing module are connected in series; the drawing acquisition module is used to acquire the target power system DWG drawing to be parsed and upload the data information to the drawing processing module; the drawing processing module is used to preprocess the acquired DWG drawing according to the received data information and upload the data information to the model building module; the model building .... The system first receives data and then uses it to construct a device symbol detection model based on deep learning and graph neural networks. This data is then uploaded to the drawing association module. The drawing association module uses the constructed model to extract symbol and parameter features from the received data, associates the parameters and symbols, and uploads the data to the topology reconstruction module. The topology reconstruction module processes the received data using graph attention networks to reconstruct the topology relationships and uploads the data to the drawing parsing module. Finally, the drawing parsing module integrates, encapsulates, and converts the received data to complete the parsing of the target power system DWG drawing.

Claims

1. An intelligent parsing method for power system DWG drawings, comprising the following steps: S1. Obtain the DWG drawing of the target power system to be parsed; S2. Preprocess the DWG drawings obtained in step S1; S3. Construct a device symbol detection model based on deep learning and graph neural networks; S4. Using the model constructed in step S3, extract symbolic and parameter features from the data information obtained in step S2, and associate the parameters and symbols. S5. Based on the graph attention network, process the data information obtained in step S4 to reconstruct the topological relationships; S6. Integrate, encapsulate, and convert the data information obtained in step S5 to complete the parsing of the target power system DWG drawing.

2. The intelligent parsing method for power system DWG drawings according to claim 1, characterized in that... Step S2, which involves preprocessing the DWG drawing obtained in step S1, specifically includes the following steps: Load DWG drawings; Traverse all entities in the model space and analyze the geometric primitive features and non-geometric primitive features of the entity model; For a defined block in a spatial entity, the original primitives included in the defined block are parsed to determine the geometric primitive features and non-geometric primitive features in the defined block; All the entity information obtained from the parsing is stored in a structured data container to obtain a structured vector dataset; where each record represents a primitive and each record includes all geometric attribute information and non-geometric data information. Based on the obtained structured vector dataset, generate structured vector data for image geometric calculations and topological relationship construction; A raster image is generated based on the obtained structured vector data.

3. The intelligent parsing method for power system DWG drawings according to claim 2, characterized in that... In the process of analyzing geometric primitive features: for lines, extract the 3D coordinates of the starting point and the ending point; for circles, extract the 3D coordinates of the center and the radius; for arcs, extract the 3D coordinates of the center, the radius, the starting angle, and the ending angle; for polylines, extract the 3D coordinates of all vertices and the closing symbol; for blocks, extract the name of the corresponding block, the position of the symbol in the drawing, the scaling ratio, and the rotation angle; for text, extract the string, the 3D coordinates of the insertion point in the text, the text height, and the layer it belongs to. In the parsing of non-geometric primitive features: for layers, extract the name of the layer to which the entity belongs; for colors, extract the color index or true color value of the entity; for line types, extract the line type name. The following steps are used to generate structured vector data for image geometric calculations and topological relation construction: Data cleaning and filtering: Based on the extracted layer and entity type information, the vector data is filtered to remove entities on the Defpoints layer or irrelevant annotation layers; Geometric data standardization: unify all coordinates to a unified coordinate system, normalize the rotation angle of the text, and calculate the bounding box of each device symbol; Topology construction preparation: Based on the block definition of device symbols, the positions of corresponding electrical connection points or ports are predefined or learned, and the wire primitives are preprocessed; the preprocessing includes discretizing continuous LWPOLYLINE into a sequence of line segments; Data output: Output structured vector data; The following steps are used to generate a raster image: Render View: Select the range of the model space for rendering the DWG file or specify the layout, calculate the bounding boxes of all entities to determine the area that needs to be rendered; Set rendering parameters: Set the resolution to the set value; calculate the pixel size of the image based on the rendering area and the set resolution; Perform rendering: Using the rendering engine that is compatible with the parsing library, generate the corresponding raster image based on the set rendering parameters; Metadata association: Records rendering parameters, outputs raster images and their corresponding mappings to vector coordinates.

4. The intelligent parsing method for power system DWG drawings according to claim 3, characterized in that... Step S3, which describes building a device symbol detection model based on deep learning and graph neural networks, specifically includes the following steps: Constructing a CNN branch based on ResNet: The top-level global average pooling layer and fully connected layer in the ResNet convolutional neural network are removed, and the remaining ResNet network is used as a CNN branch; the CNN branch is used to extract local detail features of device symbols; The Transformer architecture is adopted, and Transformer branches are constructed. Transformer branches are used to capture the global layout of the input image and the long-distance dependencies between symbols. The outputs of the CNN branch and the Transformer branch are added element-wise, and then passed through a... The convolutional layers are processed to obtain the output of the device symbol detection model.

5. The intelligent parsing method for power system DWG drawings according to claim 4, characterized in that... Step S4, using the model constructed in step S3, extracts symbolic and parameter features from the data information obtained in step S2 and associates the parameters and symbols, specifically including the following steps: The raster image obtained in step S2 is input into the model constructed in step S3 to obtain the device symbol detection result output by the model. For each detected device symbol, calculate the bounding box, category label, and confidence score; for each detected text, perform text recognition and output the text content and bounding box. Unify the bounding boxes of device symbols and text bounding boxes to the set coordinate system and perform associated calculations; For each device symbol, based on the association calculation results, select the text corresponding to the text box with the highest association score as the main parameter of that device symbol; at the same time, it is set that a text can only be assigned to one device symbol.

6. The intelligent parsing method for power system DWG drawings according to claim 5, characterized in that... Step S5, based on the graph attention network, processes the data information obtained in step S4 to reconstruct the topological relationships, specifically including the following steps: Convert the geometric information of the DWG drawing into a graph data structure; Based on geometric information, all possible connection relationships are obtained by filtering. All possible connections are processed using a graph attention network to obtain the connection probabilities of different importance between different nodes in the graph data structure. Based on the obtained connection probabilities, a connection list is constructed; the connection list includes all devices in the DWG drawing, as well as the electrical connections between all devices.

7. The intelligent parsing method for power system DWG drawings according to claim 6, characterized in that... Step S6 involves integrating, encapsulating, and converting the data information obtained in step S5 to complete the parsing of the target power system DWG drawing. Specifically, this includes the following steps: All the obtained data is integrated into a defined data structure; all the data includes symbol identification results, parameter extraction results, symbol connection relationships, and symbol-parameter association relationships. The obtained data results are mapped to a standard data format to ensure data universality; Complete the analysis of the target power system DWG drawings.

8. A system for implementing the intelligent parsing method for power system DWG drawings as described in any one of claims 1 to 7, characterized in that... It includes a drawing acquisition module, a drawing processing module, a model building module, a drawing association module, a topology reconstruction module, and a drawing parsing module; these modules are connected in series. The drawing acquisition module acquires the target power system DWG drawing to be parsed and uploads the data to the drawing processing module. The drawing processing module preprocesses the acquired DWG drawing based on the received data and uploads the data to the model building module. The model building module is used to build a device symbol detection model based on deep learning and graph neural networks according to the received data information, and upload the data information to the drawing association module; The drawing association module is used to extract symbol and parameter features from the received data information using the constructed model, associate the parameters and symbols, and upload the data information to the topology reconstruction module. The topology reconstruction module is used to process the received data information based on the graph attention network to reconstruct the topology relationship, and upload the data information to the drawing parsing module. The drawing parsing module is used to integrate, encapsulate, and convert the received data information to complete the parsing of the target power system DWG drawing.

Images