Artificial intelligence-based dispatch automation system graph model conversion system
Patent Information
- Application Number
- CN202211228013.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-09
- Publication Date
- 2026-08-28
- Estimated Expiration
- 2042-10-09
AI Technical Summary
[0002]每当有新电站接入或者保供电任务等需求时，自动化运维人员需要根据调度、方式发布的CAD图纸，在调度自动化系统内绘制对应的接线图并关联数据以备调度员日常监盘及调控使用，由于调度自动化系统图形界面的格式特殊(格式为dfg)，无法使用主流图形文件格式(dwg)，所以现在的自动化运维人员往往需要对照着图纸在系统上一线、一字手工绘制完成后才能发布，枯燥、重复、耗时长，工作效率低且很难确保不出现失误
[0019]本发明采用人工智能算法对以往电气图进行学习训练，生成图形转换模型，利用软件编程技术，直接导入用户的CAD图形文件，用软件CAD图形文件的数据及格式，自动生成SVG数据文件，再根据添加部分东方电子部分自定义标识，最终生成被软件可直接识别并使用的DFG数据文件，而利用该智能模型可将调度、方式下发的电气接线图快速、准确的转换为调度自动化系统特殊格式的图形来进行发布，提高自动化日常运维效率。
Smart Images
Figure CN115510002B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer technology, and more specifically to a graph conversion system for scheduling automation systems based on artificial intelligence. Background Technology
[0002] Whenever there is a need for a new power station to be connected or for power supply guarantee tasks, automation maintenance personnel need to draw the corresponding wiring diagrams and link the data in the dispatch automation system based on the CAD drawings issued by the dispatching authority. This is for the dispatcher's daily monitoring and control. Because the graphical interface of the dispatch automation system has a special format (dfg), it cannot use mainstream graphic file formats (dwg). Therefore, automation maintenance personnel often have to manually draw the diagrams line by line on the system before publishing them. This process is tedious, repetitive, time-consuming, inefficient, and makes it difficult to ensure error-free operation.
[0003] As shown in the diagrams, when the power supply network involves voltages from 500kV to 380V and is extremely complex, it often takes four working days to complete, resulting in very low efficiency. This makes it impossible to quickly respond to power supply guarantee tasks or the commissioning of new substations. Moreover, the dispatch automation system has very high requirements for the accuracy of the wiring diagrams, and it is difficult to ensure that there are no errors when drawing them manually. This tedious and repetitive work is outdated. Therefore, we propose a diagram conversion system for dispatch automation systems based on artificial intelligence. Summary of the Invention
[0004] To address the aforementioned shortcomings, the technical problem this invention aims to solve is to provide an AI-based scheduling automation system graph conversion system, comprising the following conversion process:
[0005] S1: Import the DWG format CAD file into the conversion system. The system extracts the CAD module parameters from the CAD file and converts the CAD module parameters.
[0006] S2: After the system extracts and converts the parameters, an SVG file is generated;
[0007] S3: Generate the DFG file after the system generates the SVG file.
[0008] In the above-mentioned technical solution of the AI-based scheduling automation system map-to-model conversion system, preferably, the CAD module parameter extraction and conversion in step S1 further includes the following process:
[0009] S1: After importing the CAD file, a calculation diagram is generated, and the system begins to calculate the custom operator;
[0010] S2: After the custom operator calculation is completed, set the file to the target format, and at the same time convert the calculation graph to the target format;
[0011] S3: After converting the computation graph to the target format, convert the computation graph to Caffe, and then convert both computation graphs to ONNX;
[0012] S4: After the computation graph format conversion is completed, inference is performed using onnxruntime and caffe to obtain end-to-end model conversion.
[0013] In the above-mentioned technical solution of the graph-model conversion system for the scheduling automation system based on artificial intelligence, preferably, the computation graph generation in step S1 mainly uses the widely used training framework PyTorch or the training framework SenseParrots. Since the dynamic graph expression is easier for users to quickly implement and iterate algorithms, the dynamic graph framework will interpret and execute the model code line by line to run the model. The essence of computation graph generation is to statically express the dynamic graph model. Currently, there are three common methods for establishing the static expression of the model: code semantic analysis; model object analysis; and model execution tracking. During the model inference process, the framework will record the operator information such as the type of the executed operator, input and output, hyperparameters, and parameters. Finally, the operator node information and model information obtained during the inference process are combined to obtain the final static computation graph.
[0014] In the above-mentioned technical solution of the scheduling automation system graph conversion system based on artificial intelligence, preferably, in the calculation of the custom operator in step S2, the code that the training framework itself cannot track and record is taken as a custom operator. The user defines the information recorded by this operator as a node in the computation graph. In actual implementation, these calculations are written into a Function or Module. Then, the user defines the information expression of the computation node corresponding to this calculation in the Function or Module. In this way, each time the defined Function or Module is called, the corresponding operator information can be recorded in the computation graph.
[0015] In the above-mentioned technical solution of the graph-model conversion system of the scheduling automation system based on artificial intelligence, preferably, the target format in step S3 is to convert the model to an intermediate format, and then the inference framework reads the intermediate format based on protobuf. The conversion of the computation graph to the target format is to parse the static computation graph and perform conversion and alignment according to the definition of the computation graph and the definition of the target format.
[0016] In the above-mentioned technical solution of the scheduling automation system graph conversion system based on artificial intelligence, preferably, the inference using onnxruntime and caffe in step S4 further includes: deployment frameworks that typically support the ONNX format are generally extended based on onnxruntime, and deployment frameworks that support the caffe format are generally extended based on native caffe. By leveraging the inference and execution capabilities of onnxruntime and caffe, the system provides the ability to run at the same operator expression level on the x86 or CUDA platform and the hardware platform.
[0017] In the above-mentioned technical solution of the scheduling automation system graph conversion system based on artificial intelligence, preferably, in the end-to-end model conversion in step S4, the TensorRT deployment framework of the CUDA platform officially produced by NVIDIA is used to directly realize the end-to-end model conversion.
[0018] As can be seen from the above technical solution, the AI-based scheduling automation system graph conversion system provided by the present invention has the following beneficial effects compared with the prior art:
[0019] This invention employs artificial intelligence algorithms to learn and train on previous electrical diagrams, generating a graphic conversion model. Using software programming technology, it directly imports the user's CAD graphic files, automatically generates SVG data files using the data and format of the CAD graphic files, and then adds custom identifiers for Dongfang Electronics, ultimately generating DFG data files that can be directly recognized and used by the software. This intelligent model can quickly and accurately convert electrical wiring diagrams issued by scheduling and mode into graphics in a special format for the scheduling automation system for publication, improving the efficiency of automated daily operation and maintenance. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments of the present invention or the prior art will be briefly introduced and explained below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a schematic diagram of the graph-to-image conversion system of the scheduling automation system based on artificial intelligence in this invention;
[0022] Figure 2 This is a schematic diagram illustrating the specific process of parameter extraction and conversion in the CAD module of the AI-based scheduling automation system's drawing-model conversion system in this invention. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0024] To provide a clearer explanation and description of the technical solution and implementation of the present invention, several preferred specific embodiments for implementing the technical solution of the present invention are described below.
[0025] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly on or indirectly set on the other component; when a component is referred to as being "connected to" another component, it can be directly connected to or indirectly connected to the other component.
[0026] Furthermore, the terms "inner" and "outer", "front" and "back", "left" and "right", "vertical" and "horizontal", "top" and "bottom" used in this document to indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings are only for the convenience of describing this application and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.
[0027] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.
[0028] Specific Implementation Example 1.
[0029] The DWG format CAD file is imported into the conversion system. The system extracts the CAD module parameters from the CAD file and converts them. After the system completes the parameter extraction and conversion, an SVG file is generated. After the system generates the SVG file, a DFG file is generated. During the CAD module parameter extraction and conversion process, after the CAD file is imported, a computational graph is generated, and the system begins calculating custom operators. After the custom operator calculation is completed, the file is set to the target format, and the computational graph is converted to the target format. After the computational graph is converted to the target format, it is converted into Caffe, and then both computational graphs are converted into ONNX. After the computational graph format conversion is completed, inference is performed using ONNX runtime and Caffe to obtain an end-to-end model conversion.
[0030] Computational graph generation primarily utilizes widely adopted training frameworks such as PyTorch or SenseParrots. Because dynamic graphs offer a more convenient way for users to quickly implement and iterate algorithms, these frameworks interpret and execute model code line by line to run the model. The essence of computational graph generation is to statically represent the dynamic graph model. Currently, there are three common methods for establishing static model representations: code semantic analysis; model object analysis; and model execution tracing. During model inference, the framework records operator information such as the type of the executed operator, input / output, hyperparameters, and parameters. Finally, the operator node information obtained during inference is combined with the model information to obtain the final static computational graph.
[0031] In the computation of custom operators, code that the training framework itself cannot track and record is treated as a custom operator. The user defines the information recorded by this operator as a node in the computation graph. In the actual implementation, these computations are written into a Function or Module. Then, the user defines the information representation of the computation node corresponding to this computation in the Function or Module. In this way, each time the defined Function or Module is called, the corresponding operator information can be recorded in the computation graph.
[0032] The target format is a conversion of the model into an intermediate format, which is then read by the inference framework. This intermediate format is based on protobuf. The conversion of the computation graph to the target format involves parsing the static computation graph and performing conversion and alignment based on the definitions of the computation graph and the target format. Inference is performed using ONNX runtime and Caffe. Specifically, deployment frameworks that support the ONNX format are usually extended based on ONNX runtime, while deployment frameworks that support the Caffe format are usually extended based on native Caffe. By leveraging the inference and runtime capabilities of ONNX runtime and Caffe, the same operator expression level can be provided on x86 or CUDA platforms as on the hardware platform.
[0033] Specific Implementation Example 2.
[0034] The DWG format CAD file is imported into the conversion system. The system extracts the CAD module parameters from the CAD file and converts them. After the system completes the parameter extraction and conversion, an SVG file is generated. After the system generates the SVG file, a DFG file is generated. During the CAD module parameter extraction and conversion process, after the CAD file is imported, a computational graph is generated, and the system begins calculating custom operators. After the custom operator calculation is completed, the file is set to the target format, and the computational graph is converted to the target format. After the computational graph is converted to the target format, it is converted into Caffe, and then both computational graphs are converted into ONNX. After the computational graph format conversion is completed, inference is performed using ONNX runtime and Caffe to obtain an end-to-end model conversion.
[0035] CAD module parameter extraction:
[0036] By importing CAD drawing files, commonly used CAD drawing components can be set as complete modules and named (e.g., M01, M02). Modules of the same component will have the same name. The drawing conversion tool will identify and extract the names of CAD drawing component modules, such as identifying names and locations.
[0037] CAD module parameter conversion:
[0038] Graphical interoperability based on Scalable Vector Graphics (SVG) / Common Information Model (CIM) in power grid dispatch automation system, that is, combining SVG and CIM to realize dynamic conversion between Energy Management System (EMS) graphic files and SVG files, improves the level of web-based graphic publishing and facilitates the export / import of graphics between different EMS, demonstrating the advantages of SVG technology in power system graphic interaction, and has been applied in practice;
[0039] If we want to display graphics on the target system, there are two methods: First, use the target system's drawing tools to draw the table and the text within the table stroke by stroke (which can be considered as SVG graphic objects); Second, use software programming technology to directly import the user's CAD drawing file, use software to read the data and format parameters of the CAD drawing file, automatically generate an SVG data file, add some custom identifiers, and finally generate a DFG data file that can be directly recognized and used by the user's software.
[0040] The drawing conversion tool will convert the identified and extracted CAD graphic component modules. Different CAD graphic component module names will be converted into circles with different size parameters. For example, the M01 module component will be converted into a circle with a 5mm radius, and the M02 module component will be converted into a circle with a 6mm radius. The tool will also display the data text boxes that it has set in advance for circles with different size parameters around the circles. The drawing conversion tool will only extract and convert the modules corresponding to the pre-set extracted CAD graphic component module names. Other CAD graphics will be converted and displayed as usual. For example, lines (including position, size, and color information) can be converted directly.
[0041] Computational graph generation primarily utilizes widely adopted training frameworks such as PyTorch or SenseParrots. Because dynamic graphs offer a more convenient way for users to quickly implement and iterate algorithms, these frameworks interpret and execute model code line by line to run the model. The essence of computational graph generation is to statically represent the dynamic graph model. Currently, there are three common methods for establishing static model representations: code semantic analysis; model object analysis; and model execution tracing. During model inference, the framework records operator information such as the type of the executed operator, input / output, hyperparameters, and parameters. Finally, the operator node information obtained during inference is combined with the model information to obtain the final static computational graph.
[0042] In the computation of custom operators, code that the training framework itself cannot track and record is treated as a custom operator. The user defines the information recorded by this operator as a node in the computation graph. In the actual implementation, these computations are written into a Function or Module. Then, the user defines the information representation of the computation node corresponding to this computation in the Function or Module. In this way, each time the defined Function or Module is called, the corresponding operator information can be recorded in the computation graph.
[0043] The target format is a conversion of the model into an intermediate format, which is then read by the inference framework. This intermediate format is based on protobuf. The conversion of the computation graph to the target format involves parsing the static computation graph and performing conversion and alignment based on the definitions of the computation graph and the target format. Inference is performed using ONNX runtime and Caffe. Specifically, deployment frameworks that support the ONNX format are usually extended based on ONNX runtime, while deployment frameworks that support the Caffe format are usually extended based on native Caffe. By leveraging the inference and runtime capabilities of ONNX runtime and Caffe, the same operator expression level can be provided on x86 or CUDA platforms as on the hardware platform.
[0044] Specific Implementation Example 3.
[0045] The DWG format CAD file is imported into the conversion system. The system extracts the CAD module parameters from the CAD file and converts them. After the system completes the parameter extraction and conversion, an SVG file is generated. After the system generates the SVG file, a DFG file is generated. During the CAD module parameter extraction and conversion process, after the CAD file is imported, a computational graph is generated, and the system begins calculating custom operators. After the custom operator calculation is completed, the file is set to the target format, and the computational graph is converted to the target format. After the computational graph is converted to the target format, it is converted into Caffe, and then both computational graphs are converted into ONNX. After the computational graph format conversion is completed, inference is performed using ONNX runtime and Caffe to obtain an end-to-end model conversion.
[0046] CAD module parameter extraction:
[0047] By importing CAD drawing files, commonly used CAD drawing components can be set as complete modules and named (e.g., M01, M02). Modules of the same component will have the same name. The drawing conversion tool will identify and extract the names of CAD drawing component modules, such as identifying names and locations.
[0048] CAD module parameter conversion:
[0049] Graphical interoperability based on Scalable Vector Graphics (SVG) / Common Information Model (CIM) in power grid dispatch automation system, that is, combining SVG and CIM to realize dynamic conversion between Energy Management System (EMS) graphic files and SVG files, improves the level of web-based graphic publishing and facilitates the export / import of graphics between different EMS, demonstrating the advantages of SVG technology in power system graphic interaction, and has been applied in practice;
[0050] If we want to display graphics on the target system, there are two methods: First, use the target system's drawing tools to draw the table and the text within the table stroke by stroke (which can be considered as SVG graphic objects); Second, use software programming technology to directly import the user's CAD drawing file, use software to read the data and format parameters of the CAD drawing file, automatically generate an SVG data file, add some custom identifiers, and finally generate a DFG data file that can be directly recognized and used by the user's software.
[0051] The drawing conversion tool will convert the identified and extracted CAD graphic component modules. Different CAD graphic component module names will be converted into circles with different size parameters. For example, the M01 module component will be converted into a circle with a 5mm radius, and the M02 module component will be converted into a circle with a 6mm radius. The tool will also display the data text boxes that it has set in advance for circles with different size parameters around the circles. The drawing conversion tool will only extract and convert the modules corresponding to the pre-set extracted CAD graphic component module names. Other CAD graphics will be converted and displayed as usual. For example, lines (including position, size, and color information) can be converted directly.
[0052] Computational graph generation primarily utilizes widely adopted training frameworks such as PyTorch or SenseParrots. Because dynamic graphs offer a more convenient way for users to quickly implement and iterate algorithms, these frameworks interpret and execute model code line by line to run the model. The essence of computational graph generation is to statically represent the dynamic graph model. Currently, there are three common methods for establishing static model representations: code semantic analysis; model object analysis; and model execution tracing. During model inference, the framework records operator information such as the type of the executed operator, input / output, hyperparameters, and parameters. Finally, the operator node information obtained during inference is combined with the model information to obtain the final static computational graph.
[0053] In the computation of custom operators, code that the training framework itself cannot track and record is treated as a custom operator. The user defines the information recorded by this operator as a node in the computation graph. In the actual implementation, these computations are written into a Function or Module. Then, the user defines the information representation of the computation node corresponding to this computation in the Function or Module. In this way, each time the defined Function or Module is called, the corresponding operator information can be recorded in the computation graph.
[0054] Generally, inference frameworks that support Caffe define additional operators on top of the native Caffe (some even modify some native Caffe components). These modifications are reflected in the `caffe.proto` file. Using such `.proto` files and `.protobuf`, a Caffe model with Mean operators can be generated. This is an ecosystem centered around the inference framework; different inference frameworks provide different `caffe.proto` files, forming their own operator definitions and constraints. We usually refer to the Caffe format defined by the inference framework as the Caffe backend. Converting the computation graph to Caffe involves distributing and mapping the operators of the computation graph to different Caffe backends. There may be one-to-many or many-to-one mapping relationships between the operators in the computation graph and the operators in Caffe. By traversing the existing list of computation graph operators, we can naturally handle one-to-many conversion mappings. However, many-to-one mapping relationships require configuring separate computation graph optimization passes for each Caffe backend to preprocess the computation graph. On the other hand, inference frameworks often do not fully support the official ONNX opsets. For ONNX models, it is often necessary to use relevant simplifiers for model preprocessing and optimization.
[0055] The target format is a conversion of the model into an intermediate format, which is then read by the inference framework. This intermediate format is based on protobuf. The conversion of the computation graph to the target format involves parsing the static computation graph and performing conversion and alignment based on the definitions of the computation graph and the target format. Inference is performed using ONNX runtime and Caffe. Specifically, deployment frameworks that support the ONNX format are usually extended based on ONNX runtime, while deployment frameworks that support the Caffe format are usually extended based on native Caffe. By leveraging the inference and runtime capabilities of ONNX runtime and Caffe, the same operator expression level can be provided on x86 or CUDA platforms as on the hardware platform.
[0056] Finally, it should be noted that the structures, proportions, sizes, etc., shown in the accompanying drawings are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which this application can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size should still fall within the scope of the technical content disclosed in this application, provided that they do not affect the effects and purposes that this application can produce.
[0057] As used herein, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a…" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0058] This invention is not limited to the above-described preferred embodiments. Anyone should know that any structural changes made under the guidance of this invention, and any technical solutions that are the same as or similar to this invention, fall within the protection scope of this invention.
Claims
1. A graph conversion method for an artificial intelligence-based scheduling automation system, characterized in that, Includes the following steps: S1: Import the DWG format CAD file into the conversion system. The system extracts the CAD module parameters from the CAD file and converts the CAD module parameters. The extraction of CAD module parameters includes: identifying and extracting pre-set and named CAD graphic element modules, the identification content including the module name and location information of the CAD graphic element modules; The conversion of CAD module parameters includes: converting the corresponding CAD graphic element module into circles with different size parameters according to the different CAD graphic element module names, and configuring preset data text boxes for circles with different size parameters around the circles; S2: After the system extracts and converts the CAD module parameters, an SVG file is generated; S3: After the system generates the SVG file, add the Dongfang Electronics custom identifier to the SVG file to generate a DFG file that can be directly recognized and used by the scheduling automation system software; The conversion method employs an artificial intelligence algorithm to learn and train on previous electrical diagrams, generating a graphical conversion model, and then uses the graphical conversion model to convert the electrical wiring diagram into a graphical file used by the dispatch automation system. The extraction and conversion of CAD module parameters in step S1 specifically includes the following steps: S11: After importing the CAD file, a calculation diagram is generated, and the system begins to calculate the custom operator; S12: After the custom operator calculation is completed, set the file to the target format and convert the calculation graph to the target format. S13: After converting the computation graph to the target format, convert the computation graph to Caffe, and then convert the computation graph to ONNX; S14: After the computation graph format conversion is completed, inference is performed using onnxruntime and caffe to obtain end-to-end model conversion.
2. The graph conversion method for an artificial intelligence-based scheduling automation system according to claim 1, characterized in that, The computation graph generation in step S11 uses the training framework PyTorch or the training framework SenseParrots. The process of generating the computation graph involves statically representing the dynamic graph model. During model inference, the training framework records the type, input, output, hyperparameters, and parameter information of the executed operators, and combines the operator node information and model information obtained during inference to obtain the final static computation graph.
3. The graph conversion method for an artificial intelligence-based scheduling automation system according to claim 1, characterized in that, The computational custom operator in step S11 includes: taking code that the training framework itself cannot track and record as a custom operator, and defining the information recorded by the user as a node in the computation graph for the custom operator; in implementation, the corresponding computation is written into a Function or Module, and the information expression of the computation node corresponding to the computation is defined in the Function or Module, so that each time the defined Function or Module is called, the corresponding operator information can be recorded in the computation graph.
4. The graph conversion method for an artificial intelligence-based scheduling automation system according to claim 1, characterized in that, The target format in step S12 is an intermediate format for the inference framework to read, and the intermediate format is implemented based on protobuf; the conversion of the computation graph to the target format includes: parsing the static computation graph, and converting and aligning it according to the definition of the computation graph and the definition of the target format.
5. The method for graph conversion in an artificial intelligence-based scheduling automation system according to claim 1, characterized in that, The inference using onnxruntime and caffe in step S14 includes: providing the ability to run on x86 or CUDA platforms at the same operator expression level as the hardware platform by leveraging the inference capabilities of onnxruntime and caffe.
6. The graph conversion method for an artificial intelligence-based scheduling automation system according to claim 1, characterized in that, The end-to-end model conversion in step S14 is implemented using the CUDA platform deployment framework TensorRT.