AI-driven Gaussian three-dimensional substation modeling optimization method and system

Abstract

The invention discloses an AI-driven Gaussian three-dimensional transformer substation modeling optimization method and system, and relates to the technical field of data processing, and the method comprises the steps: carrying out the multi-angle data collection and feature extraction of a transformer substation equipment set, and obtaining a transformer substation equipment multi-dimensional image feature set and a transformer substation equipment multi-dimensional point cloud feature set; and an AI driving algorithm channel is established, rendering modeling and optimization training are carried out on the transformer substation equipment multi-dimensional image feature set and the transformer substation equipment multi-dimensional point cloud feature set, and a Gaussian three-dimensional model of a target transformer substation is established. And constructing a multi-task learning framework, and performing prediction analysis on the real-time operation data flow of the transformer substation to obtain a transformer substation operation analysis parameter set. And introducing a reinforcement learning mechanism to carry out adaptive optimization updating on the Gaussian three-dimensional model of the target transformer substation, and generating a Gaussian three-dimensional optimization model of the transformer substation. The technical problems that in the prior art, a transformer substation three-dimensional modeling method is not high in three-dimensional model precision and fidelity, and the requirement for intelligent operation and maintenance cannot be met are solved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to an AI-driven Gaussian three-dimensional substation modeling optimization method and system. Background Art

[0002] With the development of smart grids and digital operation and maintenance, three-dimensional modeling of substations plays an increasingly important role in equipment management, fault diagnosis, and maintenance planning. Traditional three-dimensional modeling methods for substations mainly rely on manual surveying and mapping and manual modeling, which are not only time-consuming and laborious but also prone to problems such as insufficient accuracy and missing details. In recent years, with the rapid development of computer vision and deep learning technologies, it has become possible to perform automated data collection using image recognition devices and laser scanning devices. Devices such as high-resolution industrial cameras, cameras mounted on drones, and three-dimensional lidar can obtain images and point cloud data of substation equipment from multiple angles. However, the multi-angle image sets and point cloud data sets obtained usually contain a large amount of noise and redundant information, and the generated three-dimensional models cannot meet the requirements of intelligent operation and maintenance.

[0003] Therefore, in the prior art, the three-dimensional modeling method for substations has technical problems such as low accuracy and fidelity of the three-dimensional model, which cannot meet the requirements of intelligent operation and maintenance. Summary of the Invention

[0004] This application provides an AI-driven Gaussian three-dimensional substation modeling optimization method and system, which solves the technical problems in the prior art that the three-dimensional modeling method for substations has low accuracy and fidelity of the three-dimensional model and cannot meet the requirements of intelligent operation and maintenance. It realizes the construction of a high-precision and high-fidelity three-dimensional substation model. At the same time, by introducing a multi-task learning framework and a transfer learning strategy, the accuracy of operation analysis and the adaptability of the model are further improved, thereby achieving the technical effects of intelligent and efficient three-dimensional substation modeling and operation and maintenance management.

[0005] This application provides an AI-driven Gaussian three-dimensional substation modeling optimization method, and the method includes: performing multi-angle data collection and feature extraction on the substation equipment set through an image recognition device and a laser scanning device to obtain a multi-dimensional image feature set of substation equipment and a multi-dimensional point cloud feature set of substation equipment; building an AI-driven algorithm channel, where the AI-driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel, and using the AI-driven algorithm channel to perform rendering modeling and optimization training on the multi-dimensional image feature set of substation equipment and the multi-dimensional point cloud feature set of substation equipment to build a Gaussian three-dimensional model of the target substation; constructing a multi-task learning framework, obtaining a multi-task substation operation analyzer through transfer training according to the multi-task learning framework, and performing predictive analysis on the real-time operation data stream of the substation based on the multi-task substation operation analyzer to obtain a substation operation analysis parameter set; introducing a reinforcement learning mechanism based on the substation operation analysis parameter set to perform adaptive optimization and update on the Gaussian three-dimensional model of the target substation to generate a Gaussian three-dimensional optimized model of the substation.

[0006] This application also provides an AI-driven Gaussian three-dimensional substation modeling optimization system, including: a feature extraction module, which is used to perform multi-angle data collection and feature extraction on the substation equipment set through an image recognition device and a laser scanning device to obtain a multi-dimensional image feature set of substation equipment and a multi-dimensional point cloud feature set of substation equipment; a three-dimensional model construction module, which is used to build an AI-driven algorithm channel, where the AI-driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel, and using the AI-driven algorithm channel to perform rendering modeling and optimization training on the multi-dimensional image feature set of substation equipment and the multi-dimensional point cloud feature set of substation equipment to build a Gaussian three-dimensional model of the target substation; an operation analysis module, which is used to construct a multi-task learning framework, obtain a multi-task substation operation analyzer through transfer training according to the multi-task learning framework, and perform predictive analysis on the real-time operation data stream of the substation based on the multi-task substation operation analyzer to obtain a substation operation analysis parameter set; a model optimization module, which is used to introduce a reinforcement learning mechanism based on the substation operation analysis parameter set to perform adaptive optimization and update on the Gaussian three-dimensional model of the target substation to generate a Gaussian three-dimensional optimized model of the substation.

[0007] The AI-driven Gaussian three-dimensional substation modeling optimization method and system proposed in this application collect multi-angle data and extract features from the substation equipment set through an image recognition device and a laser scanning device, obtaining a multi-dimensional image feature set of substation equipment and a multi-dimensional point cloud feature set of substation equipment; build an AI-driven algorithm channel, the AI-driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel, use the AI-driven algorithm channel to perform rendering modeling and optimization training on the multi-dimensional image feature set of substation equipment and the multi-dimensional point cloud feature set of substation equipment, and build a target substation Gaussian three-dimensional model; construct a multi-task learning framework, transfer and train according to the multi-task learning framework to obtain a multi-task substation operation analyzer, and perform predictive analysis on the real-time operation data stream of the substation based on the multi-task substation operation analyzer to obtain a substation operation analysis parameter set; introduce a reinforcement learning mechanism based on the substation operation analysis parameter set to adaptively optimize and update the target substation Gaussian three-dimensional model to generate a substation Gaussian three-dimensional optimized model. It solves the technical problem that the three-dimensional modeling method of the substation in the prior art has low accuracy and fidelity of the three-dimensional model and cannot meet the requirements of intelligent operation and maintenance. It realizes the construction of a substation three-dimensional model with high accuracy and high fidelity. At the same time, by introducing a multi-task learning framework and a transfer learning strategy, it further improves the accuracy of operation analysis and the adaptability of the model, thus achieving the technical effects of intelligent and efficient substation three-dimensional modeling and operation and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the application. It should be understood that the operations in the front or below do not necessarily need to be executed precisely in sequence. On the contrary, according to needs, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0009] Figure 1 It is a schematic flowchart of the AI-driven Gaussian three-dimensional substation modeling optimization method provided by the embodiment of the present application; Figure 2 It is a schematic structural diagram of the AI-driven Gaussian three-dimensional substation modeling optimization system provided by the embodiment of the present application.

[0010] Description of the reference numerals: Feature extraction module 11, three-dimensional model construction module 12, operation analysis module 13, model optimization module 14. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented in accordance with the content of the specification. And in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the specific embodiments of the present application are specifically exemplified below.

[0012] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0013] In the following description, "some embodiments" are involved, which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The terms "first / second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0014] The embodiments of the present application provide an AI-driven Gaussian three-dimensional substation modeling optimization method and system, as Figure 1 shown, the method includes: Performing multi-angle data collection and feature extraction on the substation equipment set through an image recognition device and a laser scanning device to obtain a multi-dimensional image feature set of substation equipment and a multi-dimensional point cloud feature set of substation equipment; building an AI-driven algorithm channel, the AI-driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel, and using the AI-driven algorithm channel to perform rendering modeling and optimization training on the multi-dimensional image feature set of substation equipment and the multi-dimensional point cloud feature set of substation equipment to build a target substation Gaussian three-dimensional model.

[0015] Multi - angle data collection is performed on a set of substation equipment by an image recognition device and a laser scanning device. The set of substation equipment is a pre - set set composed of all equipment that needs to be collected, and a multi - angle image set of substation equipment and a multi - angle point cloud data set of substation equipment are obtained. The image recognition device is hardware that can capture visible - light or infrared images and perform preliminary recognition processing, such as a high - resolution industrial camera or a camera carried by a drone, etc. The laser scanning device is a 3D lidar, which obtains the three - dimensional coordinate information of the target scene by emitting laser beams and receiving reflected signals. Feature extraction is performed on the multi - angle image set of substation equipment and the multi - angle point cloud data set of substation equipment to obtain a multi - dimensional image feature set of substation equipment and a multi - dimensional point cloud feature set of substation equipment. Further, an AI - driven algorithm channel is built. The AI - driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel. 3DGS is a three - dimensional Gaussian surface construction method. It generates a smooth three - dimensional surface model by performing Gaussian weighting on feature points and surface fitting. The GAN (Generative Adversarial Network) generator is responsible for generating a more realistic three - dimensional model, and the discriminator scores and judges the generation result and returns the feedback to the generator for further improvement. The AI - driven algorithm channel is used to perform rendering modeling and optimization training on the multi - dimensional image feature set of substation equipment and the multi - dimensional point cloud feature set of substation equipment to build a Gaussian three - dimensional model of the target substation.

[0016] The method provided by the embodiment of this application further includes: performing edge recognition on the multi - angle image set of substation equipment and the multi - angle point cloud data set of substation equipment to obtain an image edge information set of substation equipment and a point cloud edge information set of substation equipment; using a Gaussian filter based on the image edge information set of substation equipment to filter and enhance the multi - angle image set of substation equipment to obtain an available multi - angle image set of substation equipment; determining a rectangular fitting rule according to the point cloud edge information set of substation equipment, and performing filtering and denoising processing within the multi - angle point cloud data set of substation equipment based on the rectangular fitting rule to obtain an available multi - angle point cloud data set of substation equipment; using a convolutional neural network to respectively perform key feature extraction on the available multi - angle image set of substation equipment and the available multi - angle point cloud data set of substation equipment to obtain the multi - dimensional image feature set of substation equipment and the multi - dimensional point cloud feature set of substation equipment.

[0017] The method of obtaining a multi-dimensional image feature set and a multi-dimensional point cloud feature set of substation equipment includes: edge recognition of the multi-angle image set of substation equipment, grayscale conversion of the images in the multi-angle image set of substation equipment, and obtaining the converted image data. Further, edge information is extracted using operators such as Canny and Sobel or deep learning models to obtain edge contour information of each device image in the multi-angle image set, and obtain a substation equipment image edge information set, which includes the main edge pixel coordinates, device contours and other information in each device image. Edge recognition is performed on the multi-angle point cloud data set of substation equipment to obtain the edges of each image in the multi-angle point cloud data set of substation equipment, and the substation equipment point cloud edge information set is obtained. Subsequently, based on the substation equipment image edge information set, a Gaussian filter is used to filter and enhance the multi-angle image set of substation equipment. Near the edge, a smaller filter kernel can be used to retain the contour as much as possible, and a larger filter kernel can be used in the non-edge area to reduce noise interference. After filtering, a usable substation equipment multi-angle image set is obtained.

[0018] Further, according to the substation equipment point cloud edge information set, according to professional and technical personnel, set rectangular fitting parameters, including: distance threshold, angle threshold, minimum point number threshold, etc., to control the accuracy of searching for rectangular edges in the point cloud, and then determine the rectangular fitting rule. Through the rectangular fitting rule, in the point cloud containing a large number of three-dimensional point coordinates, find a point set or boundary that meets the rectangular shape characteristics. Identify the rectangular or approximately rectangular part of the equipment (such as the transformer base, the facade of the switch cabinet, etc.). Use the rectangular fitting rule to search and match the point cloud in the multi-angle point cloud data set of substation equipment, filter out outliers (noise points) that do not conform to the rectangular outline or deviate from the main body of the equipment, complete filtering and denoising, and obtain an available multi-angle point cloud data set of substation equipment. In the available multi-angle point cloud data set of substation equipment, the key points on the surface or edge of the equipment are retained, and redundant data that interferes with subsequent modeling is removed. Finally, the multi-angle image set of available substation equipment is input into a pre-trained or custom-trained CNN (such as classic networks such as ResNet, VGG, or special networks designed for power equipment features). Under the action of multiple convolutional layers and pooling layers, the deep features in the image are extracted to obtain the surface texture of the equipment, the shape description of different components, the color distribution and other information, and the multi-dimensional image feature set of the substation equipment is obtained. The multi-angle point cloud dataset of available substation equipment is extracted by using a convolutional neural network, and the multi-angle point cloud dataset of available substation equipment is input into a neural network for point clouds, such as PointNet, PointNet++ or other three-dimensional CNN, to extract point cloud distribution features, obtain the geometric structure, surface information and spatial position of key components of the equipment in three-dimensional space, and obtain a multi-dimensional point cloud feature set of substation equipment.

[0019] The method provided by the embodiment of the present application further includes: determining a rectangle fitting rule according to the edge information set of the substation equipment point cloud, and setting rectangle fitting parameters based on the rectangle fitting rule; randomly selecting four points in the multi-angle point cloud data set of the substation equipment, connecting them to obtain a first fitting rectangle as the optimal rectangle solution, and performing iterative fitting comparison on the optimal rectangle solution through the rectangle fitting parameters to obtain a rectangle fitting result; performing point cloud data denoising processing on the multi-angle point cloud data set of the substation equipment based on the rectangle fitting result to obtain the available multi-angle point cloud data set of the substation equipment within a preset point cloud threshold.

[0020] Obtaining the multi - angle point cloud dataset of available substation equipment includes: determining a rectangle fitting rule according to the edge information set of the substation equipment point cloud. The rectangle fitting rule is the criteria and constraints for identifying and fitting rectangles or approximate rectangle shapes in the point cloud, and this rule can be set by professional technicians in combination with the recognition object, such as setting the maximum side length, minimum side length, angle range, side length ratio, etc. And based on the rectangle fitting rule, set rectangle fitting parameters. The rectangle fitting parameters include: the maximum number of iterations, distance threshold (used to judge the distance between a point and the edge of the fitted rectangle. For example, set the distance threshold to 5 mm, and points within 5 mm of the edge line of the fitted rectangle will be regarded as qualified points), angle threshold (used to judge the angular relationship between points. For example, set the angle threshold to ±5 degrees, and the angle of the fitted rectangle should be between 85 degrees and 95 degrees), minimum point number threshold (ensuring that the number of points in the fitting area is sufficient to exclude noise or false edges). Further, randomly select four points in the multi - angle point cloud dataset of the substation equipment as candidate points for the initial fitted rectangle. The purpose of random selection is to increase the diversity of the fitting process and avoid biases caused by fixed selection. Connect to obtain the first fitted rectangle as the optimal rectangle solution, and perform iterative fitting comparison on the optimal rectangle solution through the rectangle fitting parameters to obtain the rectangle fitting result. Exemplarily, in the point cloud dataset of the switchgear, randomly select four points A, B, C, and D and try to connect them to form a rectangle. Calculate the length ratio and angle of each side. If it meets the set fitting rule, record it as a candidate rectangle. Randomly select and evaluate multiple times within the maximum number of iterations, and finally select the rectangle that best conforms to the facade characteristics of the switchgear as the optimal solution. The iterative process is to compare the number of point cloud data with a distance less than the distance threshold for the iterative rectangle. If the number of the second fitted rectangle is greater than that of the first fitted rectangle, then replace the second fitted rectangle as the optimal solution. For each point in the point cloud dataset, calculate its distance from the edge of the rectangle fitting result. If the point is within the preset point cloud threshold, keep the point; otherwise, regard it as a noise point and remove it, thereby obtaining the multi - angle point cloud dataset of the available substation equipment. The preset point cloud threshold is a pre - set distance judgment threshold. When a point greater than the preset point cloud threshold is outside the fitting edge, and vice versa, it is inside the fitting edge.

[0021] The method provided by the embodiment of the present application further includes: based on the 3DGS algorithm channel, spatially mapping and registering the multi-dimensional image feature set of the substation equipment to the multi-dimensional point cloud feature set of the substation equipment to obtain a three-dimensional Gaussian feature point set; using a three-dimensional Gaussian function to calculate the influence range and perform Gaussian weighting processing on each feature point in the three-dimensional Gaussian feature point set to obtain a Gaussian weighted point cloud data set; performing surface fitting and smoothing processing based on the Gaussian weighted point cloud data set to obtain a preliminary three-dimensional model, and performing iterative rendering optimization on the preliminary three-dimensional model to generate a substation Gaussian three-dimensional model; based on the GAN algorithm channel, performing iterative optimization training on the substation Gaussian three-dimensional model to build a target substation Gaussian three-dimensional model.

[0022] Based on the 3DGS algorithm channel, the multi-dimensional image feature set and the multi-dimensional point cloud feature set of the substation equipment are matched and converted into the same spatial coordinate system, so that the image features and the point cloud features correspond, and the spatial mapping is completed for subsequent fusion and processing. The iterative closest point algorithm is used to align the image feature set and the point cloud feature set, and precise registration is achieved by minimizing the error between the two. After registration, the image features and the point cloud features are fused to generate a three-dimensional Gaussian feature point set, and these feature points integrate the visual and geometric information of the equipment.

[0023] Further, use a three-dimensional Gaussian function to calculate the influence range of each feature point in the three-dimensional Gaussian feature point set, that is, determine the contribution degree of the point to the surrounding area in space. At the same time, perform Gaussian weighting processing, apply the three-dimensional Gaussian function to weight each feature point, and the weight decreases with the increase of distance, ensuring that the feature point has a greater influence on its nearby area and a smaller influence on the distant area, so as to obtain a Gaussian weighted point cloud data set. For the base point cloud of a transformer, set the standard deviation of the Gaussian function to 5 millimeters. A smaller standard deviation of the Gaussian function means that the influence range is concentrated, while a larger standard deviation of the Gaussian function expands the influence range. For each feature point on the base, calculate its Gaussian weight in three-dimensional space, apply the weight to the point cloud data, and generate a smooth and continuous Gaussian weighted point cloud data set to eliminate small-range noise points caused by scanning errors.

[0024] Subsequently, based on the Gaussian weighted point cloud dataset, a three-dimensional surface model is fitted on the point cloud data through a surface fitting algorithm (Poisson surface reconstruction algorithm or marching cubes method). The three-dimensional surface model is smoothed, and a filtering algorithm is used to reduce surface irregularities and improve the visual quality of the model, thereby obtaining a preliminary three-dimensional model. The preliminary three-dimensional model is input into a rendering engine (such as Blender, Maya, etc.) for preliminary visual rendering to evaluate the appearance and structure of the model. According to the rendering results, deficiencies in the model are identified, such as missing details, non-smooth surfaces, etc., and the optimization requirements are recorded. According to the feedback, the parameters of surface fitting and smoothing processing are adjusted (such as a smaller σ value, a higher smoothing intensity), and surface fitting and model generation are performed again. The above-mentioned rendering and optimization steps are repeated. Through multiple iterations, the accuracy and details of the three-dimensional model are continuously improved until the expected high-quality effect is achieved, and a Gaussian three-dimensional model of the substation is generated. Finally, based on the GAN algorithm channel, the Gaussian three-dimensional model of the substation is iteratively optimized and trained to build a target Gaussian three-dimensional model of the substation.

[0025] The method provided by the embodiment of the present application further includes: according to the GAN algorithm channel, building a three-dimensional model generator and a three-dimensional model discriminator, using the Gaussian three-dimensional model of the substation as the input of the three-dimensional model generator for optimized generation to obtain a Gaussian optimized three-dimensional model; based on the three-dimensional model discriminator, scoring and discriminating the Gaussian optimized three-dimensional model to obtain three-dimensional model performance evaluation parameters; and feeding back the three-dimensional model performance evaluation parameters to the three-dimensional model generator for iterative tuning to obtain a target Gaussian three-dimensional model of the substation.

[0026] Building the target Gaussian three-dimensional model of the substation includes: according to the GAN algorithm channel, building a three-dimensional model generator and a three-dimensional model discriminator, using the Gaussian three-dimensional model of the substation as the input of the three-dimensional model generator for optimized generation to obtain a Gaussian optimized three-dimensional model. The Gaussian optimized three-dimensional model generated by the generator is input into the discriminator, and the discriminator scores the input model to judge its authenticity and obtains three-dimensional model performance evaluation parameters. The scoring range is usually between 0 and 1, where 1 represents a real model and 0 represents a generated model. The three-dimensional model performance evaluation parameters are fed back to the three-dimensional model generator for iterative tuning. In the first round of iteration, the score of the transformer model generated by the generator is 0.6. The generator adjusts its network parameters according to the feedback of the discriminator to optimize the detailed part. In the second round of iteration, the score of the model generated by the generator is increased to 0.75. As the number of iterations increases, the generator gradually generates high-quality models with scores approaching 1. When the score of the generated model by the discriminator stabilizes at a relatively high level (such as above 0.95) and further iteration does not significantly improve the score, a target Gaussian three-dimensional model of the substation is obtained.

[0027] The method provided by the embodiments of this application further includes: building an initial generator and an initial discriminator according to the GAN algorithm channel, and determining a generator-discriminator alternating training strategy; performing alternating training and verification on the initial generator and the initial discriminator based on the generator-discriminator alternating training strategy to obtain generator loss data and discriminator loss data; performing gradient balance update on the initial generator and the initial discriminator based on the generator loss data and the discriminator loss data to obtain the 3D model generator and the 3D model discriminator.

[0028] Building the 3D model generator and the 3D model discriminator includes: building an initial generator and an initial discriminator according to the GAN algorithm channel. The GAN algorithm is a deep learning framework consisting of two parts: a generator and a discriminator. The task of the generator is to generate realistic data samples, while the task of the discriminator is to distinguish the generated samples from the real data samples. Through the adversarial training of the two, the generator continuously improves the quality of the generated data until the discriminator can no longer effectively distinguish between true and false samples. Determine the generator-discriminator alternating training strategy, that is, during the training process, the generator and the discriminator take turns to train. Specifically, first fix the generator to train the discriminator, and then fix the discriminator to train the generator to ensure that both make progress together during the confrontation.

[0029] Set the number of iterations for each training. Usually, a relatively high number of rounds can be set in the initial stage to ensure that the model has enough time to learn. Based on the generator-discriminator alternating training strategy, perform alternating training and verification on the initial generator and the initial discriminator. In each training round, record the loss data of the generator and the discriminator as a reference for model optimization. Train the discriminator: Extract a batch of real Gaussian 3D models from the real dataset as real samples. The generator generates a batch of Gaussian optimized 3D models as fake samples. Input the real samples and the fake samples into the discriminator respectively, and calculate the discrimination probability of the discriminator for the real samples and the discrimination probability for the fake samples. According to the discrimination results of the real samples and the fake samples, calculate the loss function of the discriminator (such as binary cross-entropy loss). Through the backpropagation algorithm, update the weight parameters of the discriminator to improve its ability to distinguish between real and fake samples. Train the generator: The generator generates a batch of Gaussian optimized 3D models as fake samples. Input the fake samples into the discriminator to obtain the scores of the discriminator for these samples. Calculate the loss function of the generator, and the goal is to maximize the scores of the discriminator for the fake samples, that is, to make the discriminator think that the fake samples are as real as possible. Through the backpropagation algorithm, update the weight parameters of the generator to improve its ability to generate more realistic samples.

[0030] Finally, based on the generator loss data and discriminator loss data, perform gradient balance updates on the initial generator and the initial discriminator, and adjust the learning rates or weight update steps of the generator and the discriminator to prevent one party from improving too quickly and causing the other party to be unable to be effectively trained, ensuring the balanced development of both during the training process. Terminate the training when the loss values of the generator and the discriminator tend to stabilize in a continuous number of rounds and the quality of the 3D model generated by the generator reaches the expected standard, to obtain the 3D model generator and the 3D model discriminator.

[0031] The method provided by the embodiments of this application further includes: constructing a multi-task learning framework, obtaining a multi-task substation operation analyzer through transfer training according to the multi-task learning framework, and performing predictive analysis on the real-time operation data stream of the substation based on the multi-task substation operation analyzer to obtain a set of substation operation analysis parameters; introducing a reinforcement learning mechanism based on the set of substation operation analysis parameters to adaptively optimize and update the Gaussian 3D model of the target substation, and generating an optimized Gaussian 3D model of the substation.

[0032] Identify multiple tasks that need to be learned simultaneously. In substation operation analysis, the tasks include equipment status monitoring, fault prediction, load prediction, etc. Construct a multi-task learning framework according to the tasks, and obtain a multi-task substation operation analyzer through transfer training according to the multi-task learning framework. Perform predictive analysis on the real-time operation data stream of the substation (such as sensor data streams of real-time current, voltage, temperature, etc.) based on the multi-task substation operation analyzer, and output corresponding predictive indicators or parameters to obtain a set of substation operation analysis parameters. The set of substation operation analysis parameters includes load change trends, equipment health score, fault risk levels, etc. Further, introduce a reinforcement learning mechanism to use the set of substation operation analysis parameters as feedback information for the current 3D model. Map the set of substation operation analysis parameters to the Gaussian 3D model of the target substation for adaptive optimization and update, so that the constructed Gaussian 3D model of the target substation can better reflect the operation analysis parameters of each device, and then generate an optimized Gaussian 3D model of the substation. This solves the technical problem in the prior art that the 3D modeling method of substations has low accuracy and fidelity of the 3D model and cannot meet the requirements of intelligent operation and maintenance. It realizes the construction of a substation 3D model with high accuracy and high fidelity. At the same time, by introducing the multi-task learning framework and transfer learning strategy, it further improves the accuracy of operation analysis and the adaptability of the model, thus achieving the technical effects of intelligent and efficient substation 3D modeling and operation and maintenance management.

[0033] The method provided by the embodiment of the present application further includes: selecting a source domain device operation analysis dataset according to the multi-task learning framework, where the source domain device operation analysis dataset includes multi-task operation analysis datasets of different types of devices; respectively analyzing and training the multi-task operation analysis datasets of different types of devices to obtain a source domain device operation analyzer set; determining a substation operation analysis target domain according to the multi-task learning framework, and performing correlation analysis based on the source domain device operation analyzer set and the substation operation analysis target domain to determine a transfer learning strategy; and performing transfer training optimization on the substation historical operation dataset and the source domain device operation analyzer set based on the transfer learning strategy to obtain the multi-task substation operation analyzer.

[0034] The obtaining of the multi-task substation operation analyzer through transfer training according to the multi-task learning framework includes: selecting a source domain device operation analysis dataset according to the multi-task learning framework. The multi-task learning framework contains an analysis framework composed of various substation device operation analyzers. The source domain device operation analysis dataset includes multi-task operation analysis datasets of different types of devices, that is, it includes multi-task operation analysis datasets of different types of devices, including multiple substation core devices such as transformers, circuit breakers, busbars, and disconnectors. The dataset of each device contains multiple related operation tasks, such as data corresponding to temperature monitoring, current load, fault detection, etc. When selecting the source domain device operation analysis dataset, it is necessary to judge that the data volume of the source domain device operation analysis data meets the preset data volume requirement, and if it does not meet the requirement, the corresponding source domain device is not selected. For different device types, a general operation analyzer is constructed. Each analyzer can process multiple operation tasks of the device. When performing analysis training, based on the recursive neural network algorithm, the model is trained based on the multi-task operation analysis data of different types of devices to obtain a source domain device operation analyzer set, and each device operation analyzer corresponds to one device. Exemplarily, for the operation analyzer of the transformer, according to the multi-task learning framework, the first few layers are shared convolutional layers, and the subsequent layers are respectively responsible for temperature prediction, current load prediction, and fault detection.

[0035] Furthermore, according to the multi-task learning framework, determine the target domain of substation operation analysis, where the target domain of substation operation analysis is all target devices that need to perform operation analysis. Since there may be some devices in the target domain of substation operation analysis that cannot build an operation analyzer due to a small amount of historical operation data. For the device categories that cannot build an operation analyzer, perform correlation analysis on the set of operation analyzers for source domain devices, obtain the corresponding trained operation analyzers, and determine the transfer learning strategy based on the trained operation analyzers. The transfer learning strategy is to make a small number of parameter adjustments to the target domain data based on the set of operation analyzers for source domain devices to adapt to new tasks. Use the shared layer of the source domain model as a feature extractor and only train the task-specific layer of the target domain. Finally, perform transfer training and optimization on the substation historical operation dataset and the set of operation analyzers for source domain devices based on the transfer learning strategy, and then obtain the operation analyzers for all devices in the target domain of substation operation analysis, and obtain the multi-task substation operation analyzer. The obtained multi-task substation operation analyzer not only has high-precision multi-task prediction ability, but also can flexibly adapt to the specific needs of different substations, providing strong technical support for the intelligent upgrade and digital operation and maintenance of modern power grids.

[0036] In the above text, reference is made to Figure 1 a detailed description of the AI-driven Gaussian three-dimensional substation modeling optimization method according to an embodiment of the present invention. Next, reference will be made to Figure 2 describe the AI-driven Gaussian three-dimensional substation modeling optimization system according to an embodiment of the present invention.

[0037] The AI-driven Gaussian three-dimensional substation modeling optimization system according to an embodiment of the present invention solves the technical problem in the prior art that the three-dimensional modeling method of substations has low three-dimensional model accuracy and realism and cannot meet the requirements of intelligent operation and maintenance. It realizes the construction of a high-precision and high-fidelity three-dimensional substation model. At the same time, by introducing the multi-task learning framework and the transfer learning strategy, it further improves the accuracy of operation analysis and the adaptability of the model, thus achieving the technical effects of intelligent and efficient three-dimensional substation modeling and operation and maintenance management. The AI-driven Gaussian three-dimensional substation modeling optimization system includes: a feature extraction module 11, a three-dimensional model construction module 12, an operation analysis module 13, and a model optimization module 14.

[0038] The feature extraction module 11 is used to collect multi-angle data and extract features from the substation equipment set through an image recognition device and a laser scanning device, and obtain a multi-dimensional image feature set of substation equipment and a multi-dimensional point cloud feature set of substation equipment.

[0039] The 3D model construction module 12 is used to build an AI-driven algorithm channel, and the AI-driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel. The AI-driven algorithm channel is used to perform rendering modeling and optimization training on the multi-dimensional image feature set and the multi-dimensional point cloud feature set of the substation equipment, and build a Gaussian 3D model of the target substation.

[0040] The operation analysis module 13 is used to construct a multi-task learning framework, obtain a multi-task substation operation analyzer through transfer training according to the multi-task learning framework, and perform predictive analysis on the real-time operation data stream of the substation based on the multi-task substation operation analyzer to obtain a set of substation operation analysis parameters.

[0041] The model optimization module 14 is used to introduce a reinforcement learning mechanism based on the set of substation operation analysis parameters to adaptively optimize and update the Gaussian 3D model of the target substation, and generate an optimized Gaussian 3D model of the substation.

[0042] Next, the specific configuration of the feature extraction module 11 will be described in detail. The feature extraction module 11 may further include: synchronously collecting images based on the elongated ceramic part, and the method includes: arranging a multi-view camera array for collecting each surface of the elongated ceramic part; triggering the multi-view camera array to collect images, and using an image registration algorithm to splice and correct the multi-view images; applying an edge detection algorithm to identify the surface contour features of the ceramic part, and using texture analysis technology to extract the surface texture features.

[0043] Next, the specific configuration of the feature extraction module 11 will be described in detail. The feature extraction module 11 may further include: obtaining the multi-dimensional image feature set and the multi-dimensional point cloud feature set of the substation equipment, including: collecting multi-angle data of the substation equipment set through an image recognition device and a laser scanning device to obtain a multi-angle image set and a multi-angle point cloud data set of the substation equipment; performing edge recognition on the multi-angle image set and the multi-angle point cloud data set of the substation equipment to obtain an image edge information set and a point cloud edge information set of the substation equipment; using a Gaussian filter based on the image edge information set of the substation equipment to filter and enhance the multi-angle image set of the substation equipment to obtain an available multi-angle image set of the substation equipment; determining a rectangular fitting rule according to the point cloud edge information set of the substation equipment, and performing filtering and denoising processing in the multi-angle point cloud data set of the substation equipment based on the rectangular fitting rule to obtain an available multi-angle point cloud data set of the substation equipment; using a convolutional neural network to respectively extract key features from the available multi-angle image set and the available multi-angle point cloud data set of the substation equipment to obtain the multi-dimensional image feature set and the multi-dimensional point cloud feature set of the substation equipment.

[0044] Next, the specific configuration of the three-dimensional model construction module 12 will be further described in detail. The three-dimensional model construction module 12 further includes: obtaining an available multi-angle point cloud data set of substation equipment, including: determining a rectangular fitting rule according to the substation equipment point cloud edge information set, and setting rectangular fitting parameters based on the rectangular fitting rule; randomly selecting four points in the multi-angle point cloud data set of substation equipment, connecting them to obtain a first fitting rectangle as the optimal rectangle solution, and performing iterative fitting comparison on the optimal rectangle solution through the rectangular fitting parameters to obtain a rectangular fitting result; performing point cloud data denoising processing on the multi-angle point cloud data set of substation equipment based on the rectangular fitting result to obtain the available multi-angle point cloud data set of substation equipment within a preset point cloud threshold.

[0045] Next, the specific configuration of the three-dimensional model construction module 12 will be described in detail. The three-dimensional model construction module 12 may further include: building a Gaussian three-dimensional model of the target substation, including: spatially mapping and registering the multi-dimensional image feature set of the substation equipment to the multi-dimensional point cloud feature set of the substation equipment based on the 3DGS algorithm channel to obtain a three-dimensional Gaussian feature point set; calculating the influence range and performing Gaussian weighting processing on each feature point in the three-dimensional Gaussian feature point set using a three-dimensional Gaussian function to obtain a Gaussian weighted point cloud data set; performing surface fitting and smoothing processing on the Gaussian weighted point cloud data set to obtain a preliminary three-dimensional model, and performing iterative rendering optimization on the preliminary three-dimensional model to generate a Gaussian three-dimensional model of the substation; performing iterative optimization training on the Gaussian three-dimensional model of the substation based on the GAN algorithm channel to build a Gaussian three-dimensional model of the target substation.

[0046] Next, the specific configuration of the three-dimensional model construction module 12 will be described in detail. The three-dimensional model construction module 12 further includes: building a Gaussian three-dimensional model of the target substation, including: building a three-dimensional model generator and a three-dimensional model discriminator according to the GAN algorithm channel, using the Gaussian three-dimensional model of the substation as the input of the three-dimensional model generator for optimization generation to obtain a Gaussian optimized three-dimensional model; performing scoring and discrimination on the Gaussian optimized three-dimensional model based on the three-dimensional model discriminator to obtain three-dimensional model performance evaluation parameters; feeding back the three-dimensional model performance evaluation parameters to the three-dimensional model generator for iterative tuning to obtain a Gaussian three-dimensional model of the target substation.

[0047] Next, the specific configuration of the operation analysis module 13 will be further described in detail. The operation analysis module 13 further includes: the three-dimensional model builder and three-dimensional model discriminator, including: building an initial generator and an initial discriminator according to the GAN algorithm channel, and determining an alternating training strategy for the generator-discriminator; performing alternating training and verification on the initial generator and the initial discriminator based on the generator-discriminator alternating training strategy to obtain generator loss data and discriminator loss data; performing gradient balance update on the initial generator and the initial discriminator based on the generator loss data and the discriminator loss data to obtain the three-dimensional model builder and the three-dimensional model discriminator.

[0048] Next, the specific configuration of the model optimization module 14 will be described in detail. The model optimization module 14 further includes: obtaining a multi-task substation operation analyzer by transfer training according to the multi-task learning framework, including: selecting a source domain device operation analysis data set according to the multi-task learning framework, where the source domain device operation analysis data set includes multi-task operation analysis data sets of different types of devices; respectively analyzing and training the multi-task operation analysis data sets of different types of devices to obtain a set of source domain device operation analyzers; determining a substation operation analysis target domain according to the multi-task learning framework, and performing correlation analysis based on the set of source domain device operation analyzers and the substation operation analysis target domain to determine a transfer learning strategy; performing transfer training and optimization on the substation historical operation data set and the set of source domain device operation analyzers based on the transfer learning strategy to obtain the multi-task substation operation analyzer.

[0049] The AI-driven Gaussian three-dimensional substation modeling optimization system provided by the embodiments of the present invention can execute the AI-driven Gaussian three-dimensional substation modeling optimization method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0050] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0051] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. An AI-driven optimization method for Gaussian three-dimensional substation modeling, characterized in that, The method includes: Performing multi-angle data collection and feature extraction on the substation equipment set through an image recognition device and a laser scanning device to obtain a multi-dimensional image feature set of substation equipment and a multi-dimensional point cloud feature set of substation equipment; Building an AI-driven algorithm channel, where the AI-driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel, and using the AI-driven algorithm channel to perform rendering modeling and optimization training on the multi-dimensional image feature set of substation equipment and the multi-dimensional point cloud feature set of substation equipment to build a Gaussian three-dimensional model of the target substation; Constructing a multi-task learning framework, obtaining a multi-task substation operation analyzer through transfer training according to the multi-task learning framework, and performing predictive analysis on the real-time operation data stream of the substation based on the multi-task substation operation analyzer to obtain a set of substation operation analysis parameters; Introducing a reinforcement learning mechanism based on the set of substation operation analysis parameters to adaptively optimize and update the Gaussian three-dimensional model of the target substation to generate an optimized Gaussian three-dimensional model of the substation.

2. The AI-driven Gaussian three-dimensional substation modeling optimization method according to claim 1, characterized in that, The obtaining of the multi-dimensional image feature set of substation equipment and the multi-dimensional point cloud feature set of substation equipment includes: Performing multi-angle data collection on the substation equipment set through an image recognition device and a laser scanning device to obtain a multi-angle image set of substation equipment and a multi-angle point cloud data set of substation equipment; Performing edge recognition on the multi-angle image set of substation equipment and the multi-angle point cloud data set of substation equipment to obtain an image edge information set of substation equipment and a point cloud edge information set of substation equipment; Using a Gaussian filter based on the image edge information set of substation equipment to filter and enhance the multi-angle image set of substation equipment to obtain an available multi-angle image set of substation equipment; According to the point cloud edge information set of substation equipment, determining a rectangular fitting rule, and performing filtering and denoising processing on the multi-angle point cloud data set of substation equipment based on the rectangular fitting rule to obtain an available multi-angle point cloud data set of substation equipment; Using a convolutional neural network to respectively extract key features from the available multi-angle image set of substation equipment and the available multi-angle point cloud data set of substation equipment to obtain the multi-dimensional image feature set of substation equipment and the multi-dimensional point cloud feature set of substation equipment.

3. The AI-driven Gaussian three-dimensional substation modeling optimization method according to claim 2, wherein The obtaining of the available multi-angle point cloud data set of substation equipment includes: According to the point cloud edge information set of substation equipment, determining a rectangular fitting rule and setting rectangular fitting parameters based on the rectangular fitting rule; Randomly selecting four points in the multi-angle point cloud data set of substation equipment, connecting them to obtain a first fitting rectangle as the optimal rectangle solution, and performing iterative fitting comparison on the optimal rectangle solution through the rectangular fitting parameters to obtain a rectangular fitting result; Performing point cloud data denoising processing on the multi-angle point cloud data set of substation equipment based on the rectangular fitting result to obtain the available multi-angle point cloud data set of substation equipment within a preset point cloud threshold.

4. The AI-driven Gaussian three-dimensional substation modeling optimization method according to claim 1, characterized in that, The building of the Gaussian three-dimensional model of the target substation includes: Based on the 3DGS algorithm channel, spatially mapping and registering the multi-dimensional image feature set of substation equipment to the multi-dimensional point cloud feature set of substation equipment to obtain a three-dimensional Gaussian feature point set; Calculate the influence range and perform Gaussian weighting processing on each feature point in the three-dimensional Gaussian feature point set using a three-dimensional Gaussian function to obtain a Gaussian weighted point cloud data set; Based on the Gaussian weighted point cloud data set, perform surface fitting and smoothing processing to obtain a preliminary three-dimensional model, and perform iterative rendering optimization on the preliminary three-dimensional model to generate a substation Gaussian three-dimensional model; Based on the GAN algorithm channel, perform iterative optimization training on the substation Gaussian three-dimensional model to build a target substation Gaussian three-dimensional model.

5. The AI-driven Gaussian three-dimensional substation modeling optimization method according to claim 4, wherein The building of the target substation Gaussian three-dimensional model includes: According to the GAN algorithm channel, build a three-dimensional model generator and a three-dimensional model discriminator, and use the substation Gaussian three-dimensional model as the input of the three-dimensional model generator for optimized generation to obtain a Gaussian optimized three-dimensional model; Based on the three-dimensional model discriminator, score and discriminate the Gaussian optimized three-dimensional model to obtain three-dimensional model performance evaluation parameters; Feed back the three-dimensional model performance evaluation parameters to the three-dimensional model generator for iterative tuning to obtain a target substation Gaussian three-dimensional model.

6. The AI-driven Gaussian three-dimensional substation modeling optimization method according to claim 5, wherein, The building of the three-dimensional model generator and the three-dimensional model discriminator includes: According to the GAN algorithm channel, build an initial generator and an initial discriminator, and determine the generator-discriminator alternating training strategy; Based on the generator-discriminator alternating training strategy, perform alternating training and verification on the initial generator and the initial discriminator to obtain generator loss data and discriminator loss data; Based on the generator loss data and the discriminator loss data, perform gradient balance update on the initial generator and the initial discriminator to obtain the three-dimensional model generator and the three-dimensional model discriminator.

7. The AI-driven Gaussian three-dimensional substation modeling optimization method according to claim 6, characterized in that, The obtaining of the multi-task substation operation analyzer through transfer training according to the multi-task learning framework includes: According to the multi-task learning framework, select a source domain device operation analysis data set, and the source domain device operation analysis data set includes multi-task operation analysis data sets of different types of devices; Analyze and train the multi-task operation analysis data sets of different types of devices respectively to obtain a source domain device operation analyzer set; According to the multi-task learning framework, determine the substation operation analysis target domain, and based on the source domain device operation analyzer set and the substation operation analysis target domain, perform correlation analysis to determine the transfer learning strategy; Based on the transfer learning strategy, perform transfer training and tuning on the substation historical operation data set and the source domain device operation analyzer set to obtain the multi-task substation operation analyzer.

8. The AI-driven Gaussian three-dimensional substation modeling optimization system is characterized in that, The system is used to execute the method according to any one of claims 1-7, and the system includes: A feature extraction module, configured to collect multi-angle data and extract features from a substation device set through an image recognition device and a laser scanning device to obtain a multi-dimensional image feature set of substation devices and a multi-dimensional point cloud feature set of substation devices; A three-dimensional model construction module for building an AI-driven algorithm channel, where the AI-driven algorithm channel includes a 3DGS algorithm channel and a GAN algorithm channel. The AI-driven algorithm channel is used to perform rendering modeling and optimization training on the multi-dimensional image feature set and multi-dimensional point cloud feature set of substation equipment, and build a Gaussian three-dimensional model of the target substation; An operation analysis module for constructing a multi-task learning framework, obtaining a multi-task substation operation analyzer through transfer training according to the multi-task learning framework, and performing predictive analysis on the real-time operation data stream of the substation based on the multi-task substation operation analyzer to obtain a set of substation operation analysis parameters; A model optimization module for introducing a reinforcement learning mechanism based on the set of substation operation analysis parameters to adaptively optimize and update the Gaussian three-dimensional model of the target substation, and generate an optimized Gaussian three-dimensional model of the substation.

Images