A 3D real-scene intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence

Abstract

A three-dimensional real-scene intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence belongs to the field of intelligent operation and maintenance of power systems. The steps are: 1. Collect substation information, determine the hierarchical relationship and category distinction of the model, and preliminarily build a visual three-dimensional real-scene platform; 2. Use LOD technology to reduce data complexity and optimize the dynamic scheduling and rendering of the model; 3. Perform evaluation and fault diagnosis, build a knowledge map of power equipment health management and operation and maintenance, and combine digital twin technology to realize monitoring and intelligent operation and maintenance of physical entities; 4. Apply rendering technology to the three-dimensional real-scene model, and integrate the artificial intelligence-assisted operation and maintenance mechanism into the platform to build a complete three-dimensional real-scene platform for visualizing typical power transmission lines of the power system. The present invention provides a three-dimensional real-scene intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence with good practical application effect and high degree of intelligence.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent operation and maintenance of power systems, and specifically relates to a three-dimensional real-scene intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence. Background Art

[0002] As the global digitalization wave surges, digital transformation is penetrating every corner of the energy sector with unprecedented depth and breadth. As the core hub of the energy supply system, the power system continues to expand in scale and become increasingly complex, creating an unprecedented demand for refined management and efficient operations and maintenance.

[0003] Traditional power operations and maintenance (O&M) methods have relied primarily on manual inspections and simple data logging and analysis. These shortcomings are evident in the face of the massive amounts of equipment operating data generated by today's power systems, the volatile grid operating conditions, and the complex and diverse geographical environments. Manual inspections are not only inefficient, time-consuming, and labor-intensive, but are also susceptible to subjective factors and natural conditions, making it difficult to promptly detect and accurately identify potential faults. Simple data logging and analysis methods also fail to fully uncover the underlying operational patterns and anomalies of equipment hidden in the vast amounts of data, resulting in a often reactive O&M process that severely impacts the reliability and stability of power system operations. Summary of the Invention

[0004] The present invention provides a three-dimensional real-scene intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence. Its purpose is to overcome the shortcomings and defects of the existing technology and provide a three-dimensional real-scene intelligent operation and maintenance method for power systems with high scientificity, strong accuracy, excellent reliability and high intelligence. It can realize high-precision status assessment, active fault prediction and safe and efficient operation and maintenance of complex transmission lines in dynamic environments, and solve the core problems of low model rendering efficiency, strong manual dependence, single interactive experience and high risk of high-risk operations in traditional technologies.

[0005] The present invention is achieved through the following technical solutions: a three-dimensional real-scene intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence, the steps are:

[0006] Step 1: Using GIM technology, we collected substation drawings and on-site photos to determine the model's hierarchical relationships and categorization. Next, we processed materials, created textures, mapped the building structure, and modeled the electrical equipment. Finally, we integrated the 3D scene and exported the constructed 3D model, essentially establishing a 3D visualization platform for a typical transmission line.

[0007] 1.1 Use the GIM model cloud platform to achieve data management.

[0008] On-site photos of transmission lines, towers, insulators, and key substation equipment are taken. Topographical data of the line corridors is obtained through drone aerial photography or satellite mapping. Equipment inventory information is compiled to provide comprehensive raw data support for 3D modeling. GIM technology standardizes power grid engineering information and enables multi-platform information collection and application based on a unified model framework and presentation format. The GIM system boasts lightweight data, a clearly structured model, and strong simulation capabilities, making it suitable for the full lifecycle management of power grid data.

[0009] 1.2 Determine the hierarchical relationship and category distinction of the model, optimize the model, integrate the 3D scene and output the constructed 3D model.

[0010] According to the reference association rules, the GIM file model is parsed to its original level and stored in a selected data structure. Data parsing encompasses three aspects: geometric models, other models, and hierarchical nesting. By planning the model's structural layering and classification logic, the model is divided into a macro-level for the overall transmission line layout, a meso-level for the substation building structure, and a micro-level for electrical equipment details. A distinction is made between a static model consisting of buildings, towers, and terrain, and a dynamic model consisting of operational electrical equipment, ensuring a clear model structure and rational classification.

[0011] The GIM model's parsing process is as follows: for files stored in various formats, the file type is first identified based on the file extension, and then the corresponding parsing method is used to read the data. The read data is then written into an intermediate data structure. Due to the differences between different file types, different processing methods must be used to ensure data accuracy and completeness.

[0012] In the GIM structure, starting from the leaf node of the spatial matrix, iterative operations are performed from bottom to top, continuously multiplying the spatial matrix of each level by the matrix of the previous level, and finally obtaining the absolute matrix Av of the model relative to the coordinate origin, where Av can be expressed as:

[0013]

[0014] Where A B1 ~A C It means that for each primitive in levels B1 to C, there is a corresponding space transformation matrix.

[0015] Using 3ds Max software for polygonal modeling, component-based modeling and parametric design methods were employed to generate standardized models. The geometric information of the power grid within the GIM model was parametrically expressed, generating 3D mesh data that was then converted to a visual model. During the modeling phase, the number of sampling points was optimized based on the model type and parameters. After modeling, a triangulated meshing algorithm was used to optimize the number of triangles on the model surface. The basic primitives within the model file can be categorized as simple or complex, and combined with other model transformations, these primitives form a parametric model. Texture mapping was used to enrich the model details and enhance the visual quality.

[0016] Step 2: Study the model simplification mechanism, use multi-level of detail (LOD) technology to reduce data complexity, optimize the dynamic scheduling and rendering of the model, ensure efficient loading and rendering of large-scale scenes, and improve the rendering efficiency and visual effects of the model.

[0017] 2.1 The application of 3D model rendering technology has significantly improved the rendering efficiency and visual quality of 3D real-life models of typical power transmission lines. LOD technology is used to reduce data complexity, thereby improving the rendering efficiency of model data.

[0018] LOD technology, namely multi-level of detail technology, its core is to make decisions on resource allocation when rendering objects based on the position and importance of the nodes of the object model in the display environment. Specifically, it is to reduce the number of faces and the degree of detail of non-important objects to achieve efficient rendering operations. Since the number of faces of the three-dimensional model is mainly concentrated in the magnet part of the back field, it is most important to use LOD technology to make the model for this part. The high-poly, medium-poly and low-poly models of the back field magnet are established respectively. The relevant parameters of different models are compared in the following table:

[0019]

[0020] Place the three models in the same position, create an empty object and add an LOD component. Select the model to be displayed according to the different distances of the camera. When the model height occupies 10% of the view, the low model is displayed; when the model height occupies 50%, the medium model is activated; when it exceeds 90%, the high model is displayed.

[0021] 2.2 Improve the visual effects and rendering efficiency of the model through 3D model rendering technology, and optimize color rendering and visual effects.

[0022] The adaptive interactive system for distribution network automation terminals is divided into three functional layers: data layer, communication layer, processing layer, interaction layer, and interface layer. The processing layer is responsible for 3D modeling and color rendering of the terminal. It constructs the 3D model through geometric, physical, and behavioral modeling. The specific steps include: building an environmental model, constructing a 3D virtual scene for the terminal, introducing control logic, and ultimately establishing a 3D model of the distribution network terminal.

[0023] In the terminal three-dimensional model, color rendering can enhance the color gradient effect and make it easier for the visual system to distinguish. There are two main problems with color gradient, one is to choose the right color, and the other is to achieve a gradual transition from one color to another. To solve the latter, a linear interpolation method is proposed. After completing the lighting and material property settings, in the color system, the color corresponding to the maximum elevation value is set to max, and the color corresponding to the minimum elevation value is set to min. Decompose these two colors into three components of red, green, and blue in the color system space, namely max_r, min_r, max_g, min_g, max_b, and min_b, and their value ranges are all in the interval [1.0, 1.0, 1.0]. Assume that the gradient color corresponding to a certain elevation value is X, and its red, green, and blue color components are X respectively. r 、X g 、X b After normalizing the maximum and minimum elevation values, the difference between them is 1. At this point, there is a floating point number P at X, and the value range of P is [0,1]. The expression of P is as follows:

[0024] (1)

[0025] Among them, z represents the total component of the color system (RGB). When the linear interpolation method is used, the calculation formula of each component is as follows:

[0026] (2)

[0027] In the Open Graphics Library, the color rendering algorithm executes as follows: First, data is input; then, the maximum z value (denoted as max_z) and the minimum z value (denoted as min_z) in the input data are found. Note that when the rendered colors are consistent, the values ​​of max_z and min_z are equal; when the rendered colors are different, the following applies:

[0028]

[0029] Among them, k r 、k g 、k b They are the red, green and blue components after interpolation respectively.

[0030] When Xz=min_z, we have: glColor4f(min_r, min_g, min_b, 1.0f).

[0031] When Xz=max_z, there is: glColor4f(max_r, max_g, max_b, 1.0f).

[0032] When min_z<Xz<max_z, there is: glColor4f(min_r+k r ×(Xz-min_z),min_g+k g ×(Xz-min_z),min_b+k b ×(Xz-min_z), 1.0f); that is, glColor4f(X r , X g , X b , 1.0f), where glColor4 represents the gradient color after rendering. To achieve colors that better reflect the human visual system, the terminal's 3D model interactively expresses content, requiring the corresponding RGB values ​​and RGB components. This flexibility is achieved by dynamically acquiring the elevation values ​​of data points. RGB components can be adjusted globally or locally, using different colors to express and distinguish different data sources for enhanced color rendering.

[0033] Step 3: Through machine learning-assisted operation and maintenance, knowledge graph construction, digital twin and edge-cloud collaborative technology applications, as well as system integration and testing, the intelligent operation and maintenance and decision-making support capabilities of the power system have been significantly improved, effectively ensuring the safe and stable operation of the power system.

[0034] 3.1 Machine Learning Assisted Operations

[0035] Anomaly detection for intelligent equipment operation and maintenance requires overcoming the multi-source heterogeneity of equipment monitoring data. Monitoring data includes sequential data (such as one-dimensional sensor data), spatial data (such as images and videos), and high-dimensional, multi-channel data collected from multiple sources. These data often suffer from insufficient sample size, a small number of abnormal samples, and missing labels. Anomaly detection for intelligent equipment operation and maintenance often faces data challenges: When samples are insufficient, a generative adversarial network (GAN) approach is used, with the generator estimating missing data and the discriminator distinguishing between observations and interpolated values. When the scarcity of abnormal samples leads to class imbalance, deformable convolutional autoencoder data augmentation can be used. When labels are missing, features are extracted from a pre-trained encoder and fed into a standardized stream classification network, which is then shared with the decoder for unsupervised detection.

[0036] 3.2 Build a knowledge graph and use digital twin technology to improve system operation and maintenance capabilities

[0037] To build a knowledge graph for the power system, we extract and structure information about equipment, anomaly types, causes, and solutions from operation and maintenance records, equipment ledgers, and fault reports, storing and managing it in a Neo4j graph database. Through reasoning, the knowledge graph provides precise decision-making recommendations for equipment operation and maintenance, enhancing the refinement and intelligence of operation and maintenance.

[0038] By building digital twin models of power equipment, synchronizing equipment operating data in real time, and integrating IoT technology for fault prediction and health management, this approach enables real-time monitoring and intelligent O&M of equipment, enabling the development of proactive maintenance strategies, reducing equipment failures and downtime, and improving the operational stability of power systems.

[0039] Digital twin models provide new data simulation, real-time interaction, and dynamic monitoring methods for traditional operation and maintenance tasks such as anomaly detection, remaining service life prediction, fault diagnosis, and operation and maintenance decision-making, making up for the shortcomings of traditional intelligent operation and maintenance methods under complex working conditions. Digital twin technology is a bridge connecting virtual and real equipment models, and at the same time provides the core driving force for intelligent operation and maintenance of equipment throughout its life cycle under the new paradigm of virtual-reality interaction. Based on the digital twin maturity level, the following levels of intelligent operation and maintenance of equipment throughout its life cycle driven by digital twins are proposed: DM1 for twin inspection, DM2 for twin prevention, DM3 for twin diagnosis, and DM4 for twin optimization. The different levels are described as follows:

[0040] Twin inspection to implement DM1: This refers to using digital twin models to describe and characterize operation and maintenance objects with high precision, simulate the behavior of operation and maintenance objects under various working conditions, combine anomaly detection technology to identify potential anomalies, and improve primary operation and maintenance capabilities.

[0041] DM2: refers to using the prediction and fault evolution capabilities of digital twin models, combined with real-time feedback from physical objects, to simulate the accelerated degradation of operation and maintenance objects, thereby improving the foresight and initiative of operation and maintenance work.

[0042] Twin Diagnosis DM3: This refers to embedding faults or specific operating conditions in digital twin models to simulate fault states or operating states under extreme conditions, enhance the generalization capabilities of intelligent operation and maintenance models, and assist in fault location and source tracing analysis.

[0043] DM4: This refers to the use of digital twin technology to digitally model and simulate physical entities, and the use of real-time interaction mechanisms to achieve dynamic iteration between virtual models and physical entities, optimize the performance of physical entities and realize intelligent management and control.

[0044] Step 4: By comprehensively utilizing 3D modeling and rendering technologies, integrating AI-assisted O&M modules, meticulously developing interactive features, and conducting comprehensive platform testing and optimization, we ultimately completed deployment and delivery, successfully creating a 3D visualization platform for typical power system transmission lines. This platform, with comprehensive functionality and superior performance, provides solid support for intelligent O&M in the power system.

[0045] The constructed 3D model was imported into the Unity rendering engine. Rendering efficiency was improved using Level of Detail (LOD) technology, instanced rendering, and dynamic scheduling optimization. A realistic lighting environment and PBR materials were set up to enhance realism, ensuring the platform's ability to efficiently render complex 3D scenes. The platform then integrated its developed intelligent operations and maintenance technology, integrating machine learning models for power generation forecasting and fault diagnosis with a knowledge graph for equipment health management. Leveraging a synchronized digital twin model, the platform enabled real-time monitoring of equipment and proactive failure prediction. This increased data processing speed and accelerated model updates, providing operators with intelligent decision-making support. Interactive features were developed, including a data visualization interface that displayed real-time equipment status, fault information, and operational recommendations.

[0046] Functional testing is conducted to verify the rendering quality of the model, the accuracy of AI decisions, and the smoothness of interactive operations. Performance testing is also implemented to evaluate rendering frame rate, data processing rate, and system response time. Based on the test results, algorithm parameters are optimized and lighting materials are adjusted to improve platform efficiency, enhance the user experience, and ensure stable and reliable system operation. The platform is placed in the cloud or on a local server to ensure that it is accessible at all times and operates stably. The result is a comprehensive, high-performance intelligent operation and maintenance platform that facilitates subsequent maintenance and upgrades, and technically safeguards the safe and stable operation of the power system.

[0047] The invention creates beneficial effects: The invention discloses a three-dimensional real-scene intelligent operation and maintenance method for power systems based on multi-level detail modeling and artificial intelligence, which relates to the field of intelligent operation and maintenance of power systems. By integrating three-dimensional models and rendering technologies, utilizing LOD technology, instanced rendering and dynamic scheduling optimization, the rendering efficiency of large-scale scenes is improved, and real lighting environments and PBR materials are set to enhance the realism and visual effects of the model, ensuring that the platform can efficiently render complex three-dimensional scenes; integrating artificial intelligence-assisted operation and maintenance modules, combining machine learning models and knowledge graphs, and synchronizing digital twin models to achieve real-time monitoring and fault prediction of equipment, optimize data processing and model update efficiency, and provide intelligent decision-making support for operation and maintenance personnel; developing interactive functions, designing data visualization interfaces, displaying equipment status, fault information and operation and maintenance suggestions in real time, supporting VR / AR devices, enhancing the user's immersive experience, ensuring that operation and maintenance personnel can operate the platform intuitively and conveniently, and improving work efficiency; verifying the stability and reliability of the platform through functional testing and performance testing, optimizing algorithm parameters and lighting materials, and improving user experience. The present invention improves the visualization and intelligent operation and maintenance capabilities of the power system, is conducive to improving the real-time monitoring and fault prediction capabilities of transmission lines and equipment, changes "passive operation and maintenance" to "active operation and maintenance", and significantly improves the safety and operating efficiency of the power system.

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them: BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a diagram of the database access architecture of the GIM model cloud platform of Example 1;

[0050] Figure 2 is a model processing flow chart of Example 1;

[0051] Figure 3 is a three-dimensional power grid modeling optimization flow chart of Example 1;

[0052] Figure 4 This is a structural diagram of the distribution network automation terminal adaptive interactive system of Example 1;

[0053] Figure 5 This is a model color rendering flow chart of Example 1;

[0054] Figure 6 This is a typical component operation and maintenance diagram driven by the digital twin technology of Example 1;

[0055] Figure 7 It is a flow chart of the three-dimensional real-scene intelligent operation and maintenance method of the power system based on multi-level detailed modeling and artificial intelligence. DETAILED DESCRIPTION

[0056] The present invention will be further described below using the accompanying drawings and examples.

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0058] The flow chart of the invention method is as follows Figure 7 shown.

[0059] Example 1: Reference Figures 1 to 6 , which is the first embodiment of the present invention, provides a three-dimensional real-scene intelligent operation and maintenance method for a power system based on multi-level detailed modeling and artificial intelligence, including:

[0060] Step 1: Using GIM technology, we collected relevant substation data, drawings, and on-site photos to determine the model's hierarchical relationships and categorization. Next, we processed materials, created textures, mapped the building structure, and modeled the electrical equipment. Finally, we integrated the 3D scene and exported the constructed 3D model, essentially establishing a 3D visualization platform for a typical transmission line.

[0061] Furthermore, we can realize 3D scene modeling based on GIM technology and combined with real data. The database access architecture diagram using GIM technology is as follows: Figure 1 Show.

[0062] 1.1 Use the GIM model cloud platform to achieve data management.

[0063] On-site photos of transmission lines, towers, insulators, and key substation equipment are taken. Topographical data of the line corridors is obtained through drone aerial photography or satellite mapping. Equipment inventory information is compiled to provide comprehensive raw data support for 3D modeling. GIM technology standardizes power grid engineering information and enables multi-platform information collection and application based on a unified model framework and presentation format. The GIM system boasts lightweight data, a clearly structured model, and strong simulation capabilities, making it suitable for the full lifecycle management of power grid data.

[0064] 1.2 Determine the hierarchical relationship and category distinction of the model, optimize the model, integrate the 3D scene and output the constructed 3D model. The 3D model processing flow chart is as follows: Figure 2 shown.

[0065] According to the reference association rules, the GIM file model is parsed to its original level and stored in a selected data structure. Data parsing encompasses three aspects: geometric models, other models, and hierarchical nesting. By planning the model's structural layering and classification logic, the model is divided into a macro-level for the overall transmission line layout, a meso-level for the substation building structure, and a micro-level for electrical equipment details. A distinction is made between static models consisting of buildings, towers, and terrain, and dynamic models consisting of operational electrical equipment, ensuring a clear model structure and rational classification.

[0066] The GIM model's parsing process is as follows: for files stored in various formats, the file type is first identified based on the file extension, and then the corresponding parsing method is used to read the data. The read data is then written into an intermediate data structure. Due to the differences between different file types, different processing methods must be used to ensure data accuracy and completeness.

[0067] In the GIM structure, starting from the leaf node of the spatial matrix, iterative operations are performed from bottom to top, continuously multiplying the spatial matrix of each level by the matrix of the previous level, and finally obtaining the absolute matrix Av of the model relative to the coordinate origin, where Av can be expressed as:

[0068]

[0069] Where A B1 ~A CIt means that for each primitive in levels B1 to C, there is a corresponding space transformation matrix.

[0070] Using 3ds Max software for polygonal modeling, component-based modeling and parametric design methods were employed to generate standardized models. The geometric information of the power grid within the GIM model was parametrically expressed, generating 3D mesh data that was then converted to a visual model. During the modeling phase, the number of sampling points was optimized based on the model type and parameters. After modeling, a triangulated meshing algorithm was used to optimize the number of triangles on the model surface. The basic primitives within the model file can be categorized as simple or complex, and combined with other model transformations, these primitives form a parametric model. Texture mapping was used to enrich the model details and enhance the visual quality.

[0071] Step 2: Study the model simplification mechanism, use multi-level of detail LOD technology to reduce data complexity, optimize the dynamic scheduling and rendering of the model, ensure the efficient loading and rendering of large-scale scenes, and improve the rendering efficiency and visual effects of the model. The 3D model optimization flow chart is as follows: Figure 3 shown.

[0072] Furthermore, LOD technology is used to optimize the 3D model, ensure efficient loading and rendering of large-scale scenes, and improve the rendering efficiency and visual effects of the model. The flow chart is as follows: Figure 3 shown.

[0073] 2.1 The application of 3D model rendering technology has significantly improved the rendering efficiency and visual quality of 3D real-life models of typical power transmission lines. LOD technology is used to reduce data complexity, thereby improving the rendering efficiency of model data.

[0074] LOD technology, namely multi-level of detail technology, its core is to make decisions on resource allocation when rendering objects based on the position and importance of the nodes of the object model in the display environment. Specifically, it is to reduce the number of faces and the degree of detail of non-important objects to achieve efficient rendering operations. Since the number of faces of the three-dimensional model is mainly concentrated in the magnet part of the back field, it is most important to use LOD technology to make the model for this part. The high-poly, medium-poly and low-poly models of the back field magnet are established respectively. The relevant parameters of different models are compared in the following table:

[0075]

[0076] Place the three models in the same position, create an empty object and add an LOD component. Select the model to be displayed according to the different distances of the camera. When the model height occupies 10% of the view, the low model is displayed; when the model height occupies 50%, the medium model is activated; when it exceeds 90%, the high model is displayed.

[0077] 2.2 Improve the visual effects and rendering efficiency of the model through 3D model rendering technology, and optimize color rendering and visual effects.

[0078] The adaptive interactive system for distribution network automation terminals is divided into three functional layers: data layer, communication layer, processing layer, interaction layer, and interface layer. The processing layer is responsible for 3D modeling and color rendering of the terminal. It constructs the 3D model through geometric, physical, and behavioral modeling. The specific steps include: building an environmental model, constructing a 3D virtual scene for the terminal, introducing control logic, and ultimately establishing a 3D model of the distribution network terminal.

[0079] In the terminal three-dimensional model, color rendering can enhance the color gradient effect and make it easier for the visual system to distinguish. There are two main problems with color gradient, one is to choose the right color, and the other is to achieve a gradual transition from one color to another. To solve the latter, a linear interpolation method is proposed. After completing the lighting and material property settings, in the color system, the color corresponding to the maximum elevation value is set to max, and the color corresponding to the minimum elevation value is set to min. Decompose these two colors into three components of red, green, and blue in the color system space, namely max_r, min_r, max_g, min_g, max_b, and min_b, and their value ranges are all in the interval [1.0, 1.0, 1.0]. Assume that the gradient color corresponding to a certain elevation value is X, and its red, green, and blue color components are X respectively. r 、X g 、X b After normalizing the maximum and minimum elevation values, the difference between them is 1. At this point, there is a floating point number P at X, and the value range of P is [0,1]. The expression of P is as follows:

[0080] (1)

[0081] Among them, z represents the total component of the color system (RGB). When the linear interpolation method is used, the calculation formula of each component is as follows:

[0082] (2)

[0083] In the Open Graphics Library, the color rendering algorithm executes as follows: First, data is input; then, the maximum z value, denoted as max_z, and the minimum z value, denoted as min_z, in the input data are found. Note that when the rendered colors are consistent, the values ​​of max_z and min_z are equal; when the rendered colors are different, the following applies:

[0084]

[0085] Among them, k r 、k g 、kb They are the red, green and blue components after interpolation respectively.

[0086] When Xz=min_z, we have: glColor4f(min_r, min_g, min_b, 1.0f).

[0087] When Xz=max_z, there is: glColor4f(max_r, max_g, max_b, 1.0f).

[0088] When min_z<Xz<max_z, there is: glColor4f(min_r+k r ×(Xz-min_z),min_g+k g ×(Xz-min_z),min_b+k b ×(Xz-min_z), 1.0f); that is, glColor4f(X r , X g , X b , 1.0f), where glColor4 is the gradient color after rendering. To obtain a color that is more in line with the human visual discrimination system, the terminal 3D model needs to set the corresponding color RGB value and RGB component through interactive expression. In order to enable it to be flexibly set, it is achieved by dynamically obtaining the elevation value of the data point. The RGB components can be adjusted completely or partially, using different colors to express and distinguish different data sources to achieve better color rendering effects. The color rendering technology flow chart is as follows Figure 5 shown.

[0089] Step 3: Through machine learning-assisted operation and maintenance, knowledge graph construction, digital twin and edge-cloud collaborative technology applications, as well as system integration and testing, the intelligent operation and maintenance and decision-making support capabilities of the power system have been significantly improved, effectively ensuring the safe and stable operation of the power system.

[0090] 3.1 Machine Learning Assisted Operations

[0091] Anomaly detection for intelligent equipment operation and maintenance requires overcoming the multi-source heterogeneity of equipment monitoring data. Monitoring data includes sequential data (such as one-dimensional sensor data), spatial data (such as images and videos), and high-dimensional, multi-channel data collected from multiple sources. These data often suffer from insufficient sample size, a small number of abnormal samples, and missing labels. Anomaly detection for intelligent equipment operation and maintenance often faces data challenges: When samples are insufficient, a generative adversarial network (GAN) approach is used, with the generator estimating missing data and the discriminator distinguishing between observations and interpolated values. When the scarcity of abnormal samples leads to class imbalance, deformable convolutional autoencoder data augmentation can be used. When labels are missing, features are extracted from a pre-trained encoder and fed into a standardized stream classification network, which is then shared with the decoder for unsupervised detection.

[0092] 3.2 Build a knowledge graph and use digital twin technology to improve system operation and maintenance capabilities

[0093] To build a knowledge graph for the power system, we extract and structure information about equipment, anomaly types, causes, and solutions from operation and maintenance records, equipment ledgers, and fault reports, storing and managing it in a Neo4j graph database. Through reasoning, the knowledge graph provides precise decision-making recommendations for equipment operation and maintenance, enhancing the refinement and intelligence of operation and maintenance.

[0094] By building digital twin models of power equipment, synchronizing equipment operating data in real time, and integrating IoT technology for fault prediction and health management, this approach enables real-time monitoring and intelligent O&M of equipment, enabling the development of proactive maintenance strategies, reducing equipment failures and downtime, and improving the operational stability of power systems.

[0095] Digital twin models provide new data simulation, real-time interaction, and dynamic monitoring methods for traditional maintenance tasks such as anomaly detection, remaining service life prediction, fault diagnosis, and maintenance decision-making, making up for the shortcomings of traditional intelligent maintenance methods under complex working conditions. Digital twin technology is a bridge connecting virtual and real equipment models, and at the same time provides the core driving force for intelligent maintenance of equipment throughout its life cycle under the new paradigm of virtual-real interaction. The following figure shows the operation and maintenance of a typical component driven by digital twin technology: Figure 6 Based on the digital twin maturity level, the following levels of intelligent operation and maintenance of equipment throughout its life cycle driven by digital twins are proposed: DM1 for twin inspection, DM2 for twin prevention, DM3 for twin diagnosis, and DM4 for twin optimization. The different levels are described as follows:

[0096] Twin inspection to implement DM1: This refers to using digital twin models to describe and characterize operation and maintenance objects with high precision, simulate the behavior of operation and maintenance objects under various working conditions, combine anomaly detection technology to identify potential anomalies, and improve primary operation and maintenance capabilities.

[0097] DM2: refers to using the prediction and fault evolution capabilities of digital twin models, combined with real-time feedback from physical objects, to simulate the accelerated degradation of operation and maintenance objects, thereby improving the foresight and initiative of operation and maintenance work.

[0098] Twin Diagnosis DM3: This refers to embedding faults or specific operating conditions in digital twin models to simulate fault states or operating states under extreme conditions, enhance the generalization capabilities of intelligent operation and maintenance models, and assist in fault location and source tracing analysis.

[0099] DM4: This refers to the use of digital twin technology to digitally model and simulate physical entities, and the use of real-time interaction mechanisms to achieve dynamic iteration between virtual models and physical entities, optimize the performance of physical entities and realize intelligent management and control.

[0100] Step 4: By comprehensively utilizing 3D modeling and rendering technology, integrating AI-assisted operation and maintenance modules, meticulously developing interactive functions, and comprehensively conducting platform testing and optimization, we finally completed deployment and delivery, successfully creating a 3D real-life visualization platform for typical power system transmission lines. This platform is fully functional and has excellent performance, providing solid and powerful support for intelligent operation and maintenance of power systems. The structure diagram of the grid adaptive interactive system is shown below. Figure 4 shown.

[0101] The constructed 3D model was imported into the Unity rendering engine, utilizing Level of Detail (LOD) technology, instanced rendering, and dynamic scheduling optimization to improve rendering efficiency. A realistic lighting environment and PBR materials were set up to enhance realism, ensuring the platform's ability to efficiently render complex 3D scenes. Next, the platform integrated its developed intelligent operations and maintenance (O&M) technology, integrating machine learning models for power generation forecasting and fault diagnosis with a knowledge graph for equipment health management. Leveraging a synchronized digital twin model, the platform enabled real-time monitoring of equipment and proactive failure prediction. This increased data processing speed and accelerated model updates, providing O&M personnel with intelligent decision-making support. Interactive features were developed, including a data visualization interface that displayed real-time equipment status, fault information, and O&M recommendations.

[0102] Example 2: In order to verify the feasibility of the three-dimensional real-scene intelligent operation and maintenance method of the power system based on multi-level detailed modeling and artificial intelligence proposed in the present invention, a newly built 220kV conventional substation was used as the research object to construct a case for simulation analysis.

[0103] The selected 500kV overhead transmission line tower is a modular single-pole loop tower with the following parameters: ground wire crossarm height of 27.9m, three-phase crossarm heights of 24.5m for phase A, 21.4m for phase B, and 17.9m for phase C, tower height of 15.5m, ground line radius of 15mm, and tower tube length of 980mm. A three-dimensional real-life intelligent operation and maintenance platform model for the power system was constructed using the method designed in this invention. This model accurately reproduces the actual transmission line scenario and includes the necessary elements of the three-dimensional scene, transmission line, tower, tower geographic information, and tower parameters, demonstrating certain feasibility.

[0104] To verify the effectiveness of the proposed three-dimensional, real-world intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence in reproducing actual grid conditions, changes in the grid topology were simulated. Considering that changes in the grid topology directly affect the power flow distribution and operational stability of the power system, the voltage, current, and power distribution in different areas after the topology change will significantly change, and the operating status of the equipment will also be adjusted accordingly. Therefore, the power flow distribution and equipment operating status after the grid topology change were used as test objects, and different areas and equipment were used as key points for analysis.

[0105] When a grid topology changes, some power loads are transmitted via newly added lines, while others are adjusted along existing lines. The power flow distribution ratio changes dynamically based on line impedance and load demand. Under known conditions, a topology change simulation is performed in a grid equivalent model to obtain voltage, current, and power distribution data for each region after the topology change. Analysis of the data shows that when the grid topology changes, the voltage in each region shows a dynamic adjustment trend. However, due to the inertial characteristics of the power system, voltage adjustment takes time. Therefore, after a short period of time, the voltage distribution gradually stabilizes.

[0106] The test results show that the voltage, current, and power distribution changes in the grid equivalent model under conditions of topological changes are consistent with reality. This suggests that the proposed three-dimensional, real-world intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence has a certain degree of feasibility. Using this model as a reference, it can guide the development of smart grid demonstration projects, recreating the real-world grid environment. This will help improve the real-time visibility and controllability of grid topological changes, optimize the operational efficiency of the power system, provide technical support for the operation and maintenance of grid equipment, and transform "passive" operation and maintenance into "active" operation and maintenance, thereby improving the safety and reliability of grid operations.

[0107] The specific embodiments of the present invention are not exhaustive and do not constitute a limitation on the scope of protection of the claims. Those skilled in the art can conceive of other substantially equivalent alternatives based on the inspiration gained from the embodiments of the present invention without creative work, and all of them are within the scope of protection of the present invention.

Claims

1. A three-dimensional real-scene intelligent operation and maintenance method for power systems based on multi-level detailed modeling and artificial intelligence, characterized by: Step 1: Use GIM technology to collect substation drawings and on-site photos, determine the model's hierarchical relationships and category distinctions; perform material processing, texture creation, architectural structure drawing, and electrical equipment modeling; and finally integrate the 3D scene and output the constructed 3D model, initially building a visual 3D real-life platform for typical transmission lines. Step 2: Develop a model simplification mechanism, use multi-level of detail (LOD) technology to reduce data complexity, optimize the dynamic scheduling and rendering of the model, enable the loading and rendering of large-scale scenes, and ensure the rendering efficiency and visual effects of the model; 2.1 Use LOD multi-level of detail technology to reduce data complexity: Make decisions on resource allocation when rendering objects based on the location and importance of the nodes of the object model in the display environment, and reduce the number of faces and level of detail of non-important objects; 2.2 Improve the visual effects and rendering efficiency of the model through 3D model rendering technology, and optimize color rendering and visual effects; The distribution network automation terminal adaptive interactive system is divided into data layer, communication layer, processing layer, interaction layer and interface layer according to each functional module; The processing layer is responsible for 3D modeling and color rendering of the terminal. It constructs the 3D model through geometric, physical, and behavioral modeling. The steps include: building an environmental model, constructing a 3D virtual scene for the terminal, introducing control logic, and finally establishing a 3D model of the distribution network terminal. In the terminal 3D model, the color gradient effect is enhanced by linear interpolation: after completing the lighting and material property settings, in the color system, the color corresponding to the maximum elevation value is set to max, and the color corresponding to the minimum elevation value is set to min. These two colors are decomposed into three components of red, green, and blue in the color system space, namely max_r, min_r, max_g, min_g, max_b, and min_b, and their value ranges are all in the interval [1.0, 1.0, 1.0]. Let the gradient color corresponding to a certain elevation value be X, and its red, green, and blue color components are X respectively. r 、X g 、X b , after normalizing the maximum and minimum elevation values, the difference between the two is 1. There is a floating point number P at X, and the value range of P is [0,1]. The expression of P is as follows: （1） Among them, z represents the total RGB component of the color system. When the linear interpolation method is used, the calculation formula of each component is as follows: （2） In the Open Graphics Library, the execution process of the color rendering algorithm is as follows: first, input the data; then, find the maximum z value in the input data, denoted as max_z, and the minimum z value, denoted as min_z; when the rendering colors are the same, the values ​​of max_z and min_z are equal; when the rendering colors are different, there is （3） Among them, k r 、k g 、k b They are the red, green and blue components after interpolation processing respectively; When Xz=min_z, there is: glColor4f(min_r, min_g, min_b, 1.0f); When Xz=max_z, there is: glColor4f(max_r, max_g, max_b, 1.0f); When min_z<Xz<max_z, there is: glColor4f(min_r+k r ×(Xz-min_z),min_g+k g ×(Xz-min_z),min_b+k b ×(Xz-min_z), 1.0f); that is, glColor4f(X r , X g , X b , 1.0f), where glColor4 is the gradient color after rendering. To obtain colors that conform to the human visual discrimination system, the terminal 3D model uses interactive expression content to dynamically obtain the elevation value of the data point to set the corresponding color RGB value and RGB component, using different colors to express and distinguish different data sources. Step 3: Use machine learning technology to predict power generation and power consumption, assess equipment health, and diagnose faults. Build a knowledge graph for power equipment health management and operation and maintenance, and combine digital twin technology to monitor and intelligently operate physical objects. Step 4: Apply rendering technology to the 3D real-life model and integrate the AI-assisted operation and maintenance mechanism into the platform to build a complete 3D real-life visualization platform for typical power system transmission lines.

2. The method for three-dimensional real-scene intelligent operation and maintenance of a power system based on multi-level detailed modeling and artificial intelligence according to claim 1 is characterized by: The specific steps in step 1 are: 1.1 Using the GIM model cloud platform to achieve data management: Take on-site photos of transmission lines, towers, insulators, and substation equipment, and obtain topographic data of the line corridor through drone aerial photography or satellite maps. Organize equipment inventory information to provide comprehensive raw data support for 3D modeling. Use GIM to standardize power grid engineering information, and conduct multi-platform information collection and application based on a unified model framework and presentation format. 1.2 Determine the hierarchical relationship and category distinction of the model, optimize the model, integrate the 3D scene and output the constructed 3D model: According to the reference association rules, the GIM file model is parsed to the original level and a data structure is selected for storage. Data parsing includes three aspects: geometric model, other models, and hierarchical structure nesting. By planning the structural layering and classification logic of the model, the model is divided into a macro level for the overall layout of the transmission line, a meso level for the substation building structure, and a micro level for the details of the electrical equipment. It also distinguishes between static models consisting of buildings, towers, and terrain, and dynamic models consisting of operational electrical equipment. By using 3ds Max and Blender software for polygonal modeling, the geometric information model of the power grid in the GIM model is parametrically expressed to form three-dimensional grid data, which is then converted into a visual model.

3. The method for three-dimensional real-scene intelligent operation and maintenance of a power system based on multi-level detailed modeling and artificial intelligence according to claim 1 is characterized by: The specific method in step 3) is: 3.1 Machine Learning Assisted Operation and Maintenance: By collecting multi-dimensional historical operating data on power generation, power consumption, and equipment status in the power system for model training, we can predict power generation and power consumption and diagnose equipment failures. 3.2 Building a knowledge graph: The power system knowledge graph is constructed to extract information about equipment, anomaly types, causes, and solutions from operation and maintenance records, equipment ledgers, and fault report data. All of this information is structured and stored and managed in a graph database. The knowledge graph uses its reasoning capabilities to provide decision-making recommendations for equipment operation and maintenance. 3.3 Application of digital twin technology: Digital twin technology builds digital models for power equipment, synchronizes operating data in real time through the Internet of Things, predicts faults, assesses health status, and realizes real-time monitoring and intelligent operation and maintenance of equipment. The operation and maintenance team formulates proactive maintenance strategies to reduce failures and downtime.

4. The method of three-dimensional real-scene intelligent operation and maintenance of a power system based on multi-level detailed modeling and artificial intelligence according to claim 1 is characterized by: The specific method in step 4) is: 4.1 Integrate 3D model and rendering technology: First, import the 3D model generated in step 1 into the Unity rendering engine. Use the LOD technology, instanced rendering, and dynamic scheduling mechanism in step 2 to optimize rendering efficiency. Set point light sources, spotlights, and directional lights to simulate real lighting, and use PBR materials to enhance realism. 4.2 Integrate AI-assisted operation and maintenance mechanisms and develop interactive functions: Integrate machine learning models, knowledge graphs, and digital twin models into the platform to provide intelligent decision support for power generation and consumption forecasting, fault diagnosis, and equipment health assessment, and combine knowledge graph reasoning with real-time digital twin monitoring. Develop scene roaming, device operation, and data display functions. Users can operate in three-dimensional scenes through VR devices or wearable controllers and view device information in real time.

Images