Remote investigation and analysis method, system and equipment for power distribution room and medium

Abstract

The invention relates to the technical field of information, and discloses a remote investigation analysis method, system and device for a power distribution room and a medium, and the method comprises the steps: obtaining the video monitoring data and laser radar data of the power distribution room; performing target detection and feature extraction on the video monitoring data through a convolutional neural network model to obtain a target position and a target feature; based on the video monitoring data and the laser radar data, a deep learning algorithm and a reinforcement learning algorithm are adopted to carry out spatial analysis and layout judgment on the power distribution room, and a target analysis and judgment result is obtained; fusing the target position, the target feature and the target analysis and judgment result, inputting the fused result into a deep learning model based on a multi-layer perceptron for processing, and obtaining an optimal configuration strategy of the switch cabinet as a remote investigation analysis result for output; the power distribution room is remotely investigated and analyzed by integrating various technologies and algorithms, so that the generation efficiency and quality of the switch cabinet configuration strategy are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular, to a remote survey and analysis method, system, device and medium for a distribution substation. Background Art

[0002] With the continuous development of digital technology, planners configure switch cabinets in the distribution substation to adjust the low-voltage load of the distribution network, and then build a digital power grid.

[0003] At present, when planners carry out low-voltage load adjustment of the distribution network, they mostly rely on experience to go to the site to survey and analyze the switch cabinets in the distribution substation and then reconfigure them. However, this method has a low level of intelligence, takes a long time, and has low planning efficiency. At the same time, due to the differences in the technical levels of planners, the adjustment effects are uneven, resulting in problems such as low power supply reliability, poor power supply quality, and investment waste from time to time.

[0004] It can be seen that how to improve the efficiency of survey and analysis of the distribution substation and the efficiency and quality of the existing switch cabinet configuration method has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the Invention

[0005] The present invention provides a remote survey and analysis method, system, device and medium for a distribution substation, and solves the problem of how to improve the efficiency of survey and analysis of the distribution substation and the efficiency and quality of the existing switch cabinet configuration method.

[0006] To solve the above technical problem, the first aspect of the present invention provides a remote survey and analysis method for a distribution substation, including:

[0007] Obtain video surveillance data and lidar data of the distribution substation;

[0008] Perform object detection and feature extraction on the video surveillance data through a convolutional neural network model to obtain the target position and target features;

[0009] Based on the video surveillance data and the lidar data, use deep learning algorithms and reinforcement learning algorithms to perform spatial analysis and layout judgment on the distribution substation to obtain a target analysis and judgment result;

[0010] Fuse the target position, the target features and the target analysis and judgment result, and input them into a deep learning model based on a multi-layer perceptron for processing to obtain an optimal switch cabinet configuration strategy as the output of the remote survey and analysis result.

[0011] As a preferred solution, the obtaining of the video surveillance data of the distribution substation includes:

[0012] Collect the video streams of each area in the distribution substation in real time, and decompose the video streams into a number of consecutive image frames at a fixed frame rate, so that each image frame clearly reflects the real-time state in the distribution substation;

[0013] Perform normalization processing on each of the image frames to eliminate the differences between the image frames, and use a generative adversarial network model to perform enhancement processing on the normalized image frames to generate video monitoring data for the distribution substation.

[0014] As one of the preferred solutions, performing target detection and feature extraction on the video monitoring data through a convolutional neural network model to obtain the target position and target features, including:

[0015] Use a dataset of pre-annotated distribution substation equipment and scene images to train the Faster R-CNN model based on a convolutional neural network, and adjust the model parameters through the backpropagation algorithm during the training process to obtain a trained Faster R-CNN model; the positions and category information of targets including switch cabinets, pipe galleries, and switch rooms are annotated in the dataset;

[0016] Input the video monitoring data into the trained Faster R-CNN model for target detection, so that the trained Faster R-CNN model generates target candidate regions through the region proposal network layer, and use the classification layer of the trained Faster R-CNN model to classify and regress the target candidate regions to obtain the positions and categories of the targets in the video monitoring data;

[0017] Based on the positions and categories of the targets in the video monitoring data, perform feature extraction on the video monitoring data through the convolutional layer and pooling layer of the trained Faster R-CNN model to obtain target features.

[0018] As one of the preferred solutions, based on the video monitoring data and the lidar data, use deep learning algorithms and reinforcement learning algorithms to perform spatial analysis and layout judgment on the distribution substation to obtain a target analysis and judgment result, including:

[0019] Based on the target positions and the target features, use the MonoDETR model to perform fusion processing on the video monitoring data and the lidar data to perform spatial analysis on the distribution substation to obtain target spatial dimension data;

[0020] Input the video monitoring data into a semantic segmentation model and an intelligent agent model based on reinforcement learning for vacant position detection to perform layout judgment on the distribution substation to obtain a target position configuration result;

[0021] Optimize the target space dimension data and the target position configuration result by using a graph neural network model based on the attention mechanism and a reinforcement learning algorithm to obtain a target analysis and judgment result.

[0022] As one of the preferred solutions, based on the target position and the target features, use the MonoDETR model to fuse the video surveillance data and the lidar data to perform spatial analysis on the distribution substation to obtain target space dimension data, including:

[0023] Based on the target position and the target features, encode the video surveillance data into a feature vector through the encoder of the MonoDETR model, and predict the depth information of the target in the video surveillance data through the decoder of the MonoDETR model to generate a depth map;

[0024] Preprocess the lidar data so that the preprocessed lidar data is in the same coordinate system as the video surveillance data, and match and fuse the depth information in the preprocessed lidar data with the depth map to obtain optimized depth information;

[0025] Combine the optimized depth information with the camera parameters and pixel coordinates to quantify the target space dimension data of the distribution substation.

[0026] As one of the preferred solutions, input the video surveillance data into a semantic segmentation model and an agent model based on reinforcement learning for detecting free positions to perform layout judgment on the distribution substation to obtain a target position configuration result, including:

[0027] Input the video surveillance data into a pre-trained U-Net semantic segmentation model for processing to obtain a free area segmentation result;

[0028] Input the free area segmentation result into an agent model trained based on reinforcement learning for detecting free positions, so that the agent model judges whether the free area segmentation result is suitable for installing switch cabinets according to the trained free position judgment strategy to obtain a target position configuration result.

[0029] As one of the preferred solutions, optimize the target space dimension data and the target position configuration result by using a graph neural network model based on the attention mechanism and a reinforcement learning algorithm to obtain a target analysis and judgment result, including:

[0030] Take the targets in the distribution substation as nodes and the connection relationships between the targets as edges to construct a layout graph structure model of the distribution substation;

[0031] Process the layout diagram structure model based on the target position and the target space size data to obtain node features and edge features;

[0032] Input the node features and the edge features into a graph neural network model based on the attention mechanism for weight calculation to obtain node attention weights and edge attention weights;

[0033] Taking the electrical specifications as the constraint conditions and the maximization of space utilization rate as the optimization goal, optimize the target position configuration result, the node attention weights, and the edge attention weights through a reinforcement learning algorithm to obtain a layout recommendation scheme as the output of the target analysis and judgment result.

[0034] The second aspect of the present invention provides a remote survey and analysis system for a distribution room, including:

[0035] A data acquisition module for acquiring video surveillance data and lidar data of the distribution room;

[0036] A target determination module for performing target detection and feature extraction on the video surveillance data through a convolutional neural network model to obtain the target position and target features;

[0037] A layout analysis module for performing spatial analysis and layout judgment on the distribution room based on the video surveillance data and the lidar data by using deep learning algorithms and reinforcement learning algorithms to obtain a target analysis and judgment result;

[0038] A result output module for fusing the target position, the target features, and the target analysis and judgment result, and inputting them into a deep learning model based on a multi-layer perceptron for processing to obtain an optimal configuration strategy for switch cabinets as the output of the remote survey and analysis result.

[0039] The third aspect of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the remote survey and analysis method for a distribution room as described above.

[0040] The fourth aspect of the present invention provides a computer-readable storage medium, which includes a stored computer program. When the device where the computer-readable storage medium is located executes the computer program, it implements the remote survey and analysis method for a distribution room as described above.

[0041] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following:

[0042] (1) By integrating video surveillance, lidar data, and deep learning algorithms, intelligent monitoring and management of the distribution substation are realized, improving the operation and maintenance efficiency; using a convolutional neural network model to perform object detection and feature extraction on video surveillance data can accurately locate the target positions and target features in the distribution substation, providing a basis for subsequent analysis.

[0043] (2) Combining video surveillance data and lidar data, and using deep learning algorithms and reinforcement learning algorithms for spatial analysis and layout judgment helps optimize the layout of the distribution substation and improve space utilization; processing the fused data through a deep learning model of a multi-layer perceptron can formulate an optimal configuration strategy for switchgear to ensure the safe and stable operation of the distribution substation; through the integration of multiple technologies and algorithms for remote investigation and analysis of the distribution substation, labor costs are reduced, and the efficiency and quality of switchgear configuration are effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] To more clearly illustrate the technical solutions of the present invention, the drawings required for the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0045] Figure 1 is a flowchart of a remote investigation and analysis method for a distribution substation provided by an embodiment of the present invention;

[0046] Figure 2 is a structural diagram of a remote investigation and analysis system for a distribution substation provided by an embodiment of the present invention;

[0047] Figure 3 is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings and embodiments. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0049] In the description of this application, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first", "second", "third", etc. may explicitly or implicitly include one or more of such features. In the description of this application, unless otherwise stated, the meaning of "a plurality" is two or more.

[0050] In the description of this application, it should be noted that unless otherwise clearly defined and limited, the terms "installed", "connected", "connected to" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are only for the purpose of illustration and do not indicate or imply that the indicated system or component must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be construed as a limitation of the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0051] In the description of this application, it should be noted that unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by those skilled in the technical field to which this belongs. The terms used in the description of this invention in the specification are only for the purpose of describing specific embodiments and are not intended to limit this invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0052] In one embodiment, as Figure 1 shown, the first aspect of the present invention provides a remote investigation and analysis method for a distribution substation, including:

[0053] S1. Obtain video surveillance data and lidar data of the distribution substation;

[0054] S2. Perform object detection and feature extraction on the video surveillance data through a convolutional neural network model to obtain the object position and object features;

[0055] S3. Based on the video surveillance data and the lidar data, use deep learning algorithms and reinforcement learning algorithms to perform spatial analysis and layout judgment on the distribution substation to obtain a target analysis and judgment result;

[0056] S4. Integrate the target location, the target feature, and the target analysis and judgment result, and input them into a deep learning model based on a multi-layer perceptron for processing to obtain the optimal configuration strategy of the switchgear cabinet as the output of the remote inspection and analysis result.

[0057] Specifically, in the distribution substation, high-definition cameras (such as 4K resolution) and lidar (such as Velodyne VLP-16) are deployed to cover all its areas, including key positions such as the switchgear cabinet area and the pipe gallery entrance, so as to collect video surveillance data and lidar data in the distribution substation in real time. The time synchronization protocol is used to ensure the time alignment of the video and lidar data and then upload them to the cloud for remote inspection and analysis, eliminating the need for planners to go to the site for inspection, saving time and labor costs. Then, a convolutional neural network model, such as the Faster R-CNN (Faster Region-based Convolutional Neural Networks) model, is used to detect the targets (such as switchgear cabinets, pipe galleries, switch rooms, etc.) in the distribution substation based on the collected video surveillance data, obtaining the category and precise location of the targets. Then, based on the detected target objects, the convolutional layer and pooling layer in Faster R-CNN are used to extract their features to obtain the target features. Optionally, the target detection can also use models such as YOLOv5 (You Only Look Once version 5), and the feature extraction can also use ResNet (Residual Networks) or EfficientNet (Efficient Convolutional Network), etc. Then, the depth information of the video surveillance data and the lidar data is integrated, and deep learning algorithms (such as convolutional neural networks, recurrent neural networks, etc.) are used to perform spatial analysis on the integrated data to identify information such as the target space size in the distribution substation. Based on the spatial analysis result, a reinforcement learning algorithm is used to judge and optimize the layout position of the distribution substation to obtain the vacant positions, enabling the intelligent agent to simulate different layout schemes based on the information obtained from the analysis and optimization and evaluate their effects, and finally select the optimal layout scheme as the output of the target analysis and judgment result. Finally, the target location, the target feature, and the target analysis and judgment result are integrated to form input data containing rich information and input into a pre-trained deep learning model based on a multi-layer perceptron for processing, outputting and executing the optimal configuration strategy for the switchgear cabinet; among them, the configuration strategy includes information such as the recommended installation location of the switchgear cabinet and the required installation space.

[0058] In addition, it should be noted that each model directly applied in the present invention is pre-trained by combining transfer learning and self-supervised learning. First, it is pre-trained on a large-scale general image dataset to learn general visual features, and then fine-tuned on the image dataset of the distribution substation equipment and spatial position status. By using self-supervised tasks such as masked image modeling, the model learns the internal structure and features of the image on unlabeled data, further improving the generalization ability of the model and the understanding ability of the distribution substation scenario. In addition, other training methods can also be used according to the characteristics of the model, which are not specifically limited herein.

[0059] Through the automatic target detection and feature extraction of the distribution substation, the present invention realizes remote investigation and analysis, reducing labor costs; through the combination of deep learning algorithms and reinforcement learning algorithms, it conducts spatial analysis and layout judgment on the distribution substation, improving the space utilization rate of the distribution substation, ensuring that the equipment layout complies with safety regulations, and reducing potential safety hazards; through optimizing the layout, it effectively improves the generation efficiency and quality of the switch cabinet configuration strategy.

[0060] In one embodiment, the obtaining of the video monitoring data of the distribution substation includes:

[0061] Real-time collect the video streams of each area in the distribution substation, and decompose the video streams into a number of consecutive image frames at a fixed frame rate, so that each image frame clearly reflects the real-time state in the distribution substation;

[0062] Perform normalization processing on each of the image frames to eliminate the differences between the image frames, and use a generative adversarial network model to perform enhancement processing on the normalized image frames to generate the video monitoring data of the distribution substation.

[0063] Specifically, the present invention performs preliminary processing (such as denoising, enhancement, compression) on the collected video stream to improve the clarity of the video stream, and decomposes the preprocessed video stream into consecutive image frames at a fixed frame rate, for example, 25 frames per second, to ensure that each frame of the image can clearly reflect the real-time state in the distribution substation; then performs normalization processing on the decomposed images, maps the pixel values of the images uniformly to the range of [0, 1] to eliminate the differences caused by factors such as lighting and shooting angles between different images, and at the same time, adjusts all images to a unified size, such as 640×480 pixels, to meet the input requirements of the subsequent model; finally, uses a generative adversarial network to perform data augmentation on the normalized image frames, that is, through the adversarial training of the generator and the discriminator, generates image data that is similar to but different from the real distribution substation scene, not only expands the data set, but also can generate images in some rare abnormal states, enhances the adaptability of the model to various complex situations, and generates video surveillance data for the distribution substation. In addition, in order to expand the data set and enhance the generalization ability of the model, the augmentation operation of the image frames can also be performed by random cropping, that is, randomly cropping some areas from the original image to simulate the shooting effects from different perspectives, horizontal and vertical flipping to increase the diversity of the images, adjusting the brightness, contrast and saturation so that the model can adapt to images under different lighting conditions, etc.

[0064] Through a series of operations such as real-time collection, image frame decomposition, normalization processing, and GAN enhancement of the video streams in each area of the distribution substation, the present invention realizes the efficient processing and analysis of the video surveillance data of the distribution substation, not only improves the quality and stability of the surveillance data, but also enhances the intelligent level of the surveillance system, providing a strong guarantee for the safe operation of the distribution substation.

[0065] In one embodiment, step S2 includes:

[0066] Training the Faster R-CNN model based on a convolutional neural network using a data set of pre-annotated distribution substation equipment and scene images, and adjusting the model parameters through the backpropagation algorithm during the training process to obtain a trained Faster R-CNN model; the positions and category information of the targets including switch cabinets, cable galleries, and switch rooms are annotated in the data set;

[0067] Inputting the video surveillance data into the trained Faster R-CNN model for object detection, so that the trained Faster R-CNN model generates object candidate regions through the region proposal network layer, and classifies and regresses the object candidate regions using the classification layer of the trained Faster R-CNN model to obtain the positions and categories of the objects in the video surveillance data;

[0068] Based on the position and category of the target in the video surveillance data, feature extraction is performed on the video surveillance data through the convolutional layer and pooling layer of the trained Faster R-CNN model to obtain target features.

[0069] Specifically, in this embodiment, the Faster R-CNN model based on the convolutional neural network is used to perform target detection and feature extraction on the video surveillance data; among them, the Faster R-CNN model includes a Region Proposal Network layer (also known as the Region Proposal Network layer), a classification layer, a convolutional layer, and a pooling layer.

[0070] First, the Faster R-CNN model is trained using a pre-annotated dataset of power distribution room equipment and scene images. The dataset annotates the position and category information of targets such as switch cabinets, cable galleries, and switch rooms. During the training process, the parameters of the model are continuously adjusted through the backpropagation algorithm to enable the model to accurately identify different target objects, and a trained Faster R-CNN model is obtained.

[0071] Then, the video surveillance data obtained by real-time acquisition and processing is input into the trained Faster R-CNN model for target detection, enabling the model to generate candidate regions that may contain the target through the Region Proposal Network (RPN), and using the classification layer in the model to classify and regress these candidate regions to determine the category and precise position of the target.

[0072] Finally, for the detected target objects, the convolutional layer and pooling layer of the trained Faster R-CNN model are used to extract their features to obtain target features; for example, for a switch cabinet, features such as its shape, color, and texture can be extracted; for a cable gallery, features such as its orientation and width can be extracted, and these features will be used for subsequent spatial analysis and layout judgment.

[0073] The present invention trains the Faster R-CNN model using a pre-annotated dataset and applies it to target detection and feature extraction in video surveillance data, improving the accuracy and efficiency of target detection, realizing the automatic detection of power distribution room equipment and scenes, reducing manual intervention, and improving work efficiency.

[0074] In one embodiment, step S3 includes:

[0075] Based on the target position and the target features, the MonoDETR model is used to perform fusion processing on the video surveillance data and the lidar data to perform spatial analysis on the power distribution room and obtain target spatial dimension data;

[0076] Input the video surveillance data into a semantic segmentation model and an intelligent agent model based on reinforcement learning for detecting free spaces, so as to judge the layout of the distribution room and obtain a target position configuration result;

[0077] Optimize the target space dimension data and the target position configuration result by using a graph neural network model based on an attention mechanism and a reinforcement learning algorithm to obtain a target analysis and judgment result.

[0078] Specifically, the present invention first obtains target space dimension data based on a space dimension calculation method optimized by using a MonoDETR (Depth-aware Transformer for Monocular 3D Object Detection) model and lidar data according to the obtained target position and target features. Compared with traditional object detection models based on convolutional neural networks, it abandons the region proposal network and anchor box mechanism and directly performs object detection in an end-to-end manner, greatly simplifying the detection process and having better performance in dealing with complex scenes and small object detection, and being able to more accurately identify the dimensions of various devices in the distribution room and the spatial position conditions in the distribution room. Then, a semantic segmentation model and an intelligent agent model based on reinforcement learning are used to detect free spaces in the distribution room, which can automatically find the unoccupied spaces inside the distribution room and provide an important reference for the layout of switch cabinets. Finally, a graph neural network model based on an attention mechanism and a reinforcement learning algorithm are used to optimize the target positions in the distribution room to obtain a layout recommendation that meets electrical specifications and has a high space utilization rate, realizing the intelligence and automation of the spatial analysis and layout judgment of the distribution room, improving the accuracy and reliability of the analysis, and providing strong support for the reasonable planning and equipment management of the distribution room.

[0079] In one embodiment, based on the target position and the target features, the MonoDETR model is used to perform fusion processing on the video surveillance data and the lidar data to perform spatial analysis on the distribution room and obtain target space dimension data, including:

[0080] Based on the target position and the target features, the encoder of the MonoDETR model encodes the video surveillance data into a feature vector, and the decoder of the MonoDETR model predicts the depth information of the targets in the video surveillance data according to the feature vector to generate a depth map;

[0081] Preprocess the lidar data so that the preprocessed lidar data is in the same coordinate system as the video surveillance data, and match and fuse the depth information in the preprocessed lidar data with the depth map to obtain optimized depth information;

[0082] Combine the optimized depth information with the camera parameters and pixel coordinates to quantify the target space dimension data of the distribution room.

[0083] Specifically, in traditional space dimension calculation, it mainly relies on the triangulation principle. Based on the camera calibration parameters, it can obtain the internal parameters (such as focal length) and external parameters (such as rotation and translation matrices) of the camera, and through these parameters, combined with the pixel coordinates of the target object in the image using the triangulation formula, the space dimension of the target object can be initially calculated. For example, given the camera focal length, the pixel coordinates of the target object in the image, and the approximate distance estimate between the camera and the target object, the dimension of the target object in the actual space can be calculated through simple geometric relationships. However, the accuracy of this method is limited by the accuracy of the distance estimate and the resolution and noise interference of the image itself.

[0084] To further improve the calculation accuracy, the present invention introduces the MonoDETR model based on the Transformer architecture, which can effectively model the global information in the image based on the self-attention mechanism. Specifically, based on the target position and target features in the video surveillance data, the MonoDETR model encodes the input image into a feature vector through its encoder-decoder structure, and then predicts the depth information of the target object in the decoder. During actual operation, the MonoDETR model performs a series of convolutional and self-attention calculations on the input video surveillance data through its encoder to extract the high-level semantic encoded features of the image, and then the MonoDETR model predicts the depth value corresponding to each pixel point based on these features through its decoder, thereby generating a depth map.

[0085] Since lidar can directly obtain the three-dimensional point cloud data (i.e., lidar data) in the scene, to further improve the accuracy of dimension calculation, the present invention fuses the three-dimensional point cloud data obtained by lidar with the depth map generated by the MonoDETR model. The specific approach is to first preprocess the lidar data, remove its noise points, and convert it to the same coordinate system as the video surveillance data; then match and fuse the depth information in the three-dimensional point cloud data with the depth map generated by MonoDETR. For example, for a certain corner point of the switch cabinet, there is both the actual depth value measured in the point cloud data and the depth value predicted by the MonoDETR model. Through fusion strategies such as weighted averaging, a more accurate depth value is obtained, and then combined with the parameters and pixel coordinates of the camera that captured the video surveillance data, the more accurate length, width, and height dimensions of the target are calculated through geometric methods to output as the target space dimension data.

[0086] The present invention can utilize the complementary advantages of video surveillance data and lidar data through the MonoDETR model for precise spatial analysis; among them, the video surveillance data provides rich visual information, while the lidar data provides precise distance and depth information. Through the fusion processing of the MonoDETR model, three-dimensional reconstruction and precise measurement of the internal space of the distribution room can be achieved, so as to obtain accurate target space dimension data for subsequent reasonable equipment layout and planning.

[0087] In one embodiment, inputting the video surveillance data into a semantic segmentation model and an intelligent agent model based on reinforcement learning for detecting free positions to judge the layout of the distribution room and obtain a target position configuration result includes:

[0088] Inputting the video surveillance data into a pre-trained U-Net semantic segmentation model for processing to obtain a free area segmentation result;

[0089] Inputting the free area segmentation result into an intelligent agent model trained based on reinforcement learning for detecting free positions, so that the intelligent agent model judges whether the free area segmentation result is suitable for installing switch cabinets according to the trained free position judgment strategy to obtain a target position configuration result.

[0090] Specifically, the present invention classifies different regions in the distribution room video surveillance data image by using deep learning semantic segmentation technology to initially identify possible free areas, that is, using the U-Net (Convolutional Networks for Biomedical Image Segmentation) semantic segmentation model to perform downsampling and upsampling operations on the image through its encoder-decoder structure, and predicting the category (such as equipment area, free area, etc.) to which each pixel point belongs. In practical applications, the real-time video surveillance data image of the distribution room is input into the pre-trained U-Net semantic segmentation model, and the model outputs the category label of each pixel point, thereby obtaining a preliminary free area segmentation result.

[0091] Next, the Generative Adversarial Networks (GAN) reinforcement learning is introduced. GAN consists of a generator and a discriminator. The generator generates virtual scene images of the power distribution room under different layouts. These virtual scenes contain different equipment placements and possible free areas. The discriminator continuously learns to distinguish between the virtual scenes generated by the generator and the real power distribution room scene images. During the training process, the device position, operating state, and environmental parameters (such as temperature and humidity) can be used as the state space, and device movement (translation, rotation, etc.) can be used as the action space. A reward function is designed based on the safety distance, maintenance convenience, and heat dissipation efficiency. The deep Q-network or proximal policy optimization algorithm is used to train the intelligent agent model. Through the confrontation between the generator and the discriminator, their respective capabilities are continuously improved, making the generated virtual scenes more and more realistic, and then an intelligent agent model trained based on reinforcement learning is obtained.

[0092] In the generated virtual scene, a reinforcement learning agent is introduced to perform a simulation operation of "installing switchgear". The intelligent agent model interacts with the virtual environment and continuously adjusts its behavior strategy according to the reward signal feedback from the environment. For example, if the intelligent agent model attempts to place switchgear in a certain area of a virtual scene and the area meets the installation conditions (such as sufficient space, no conflict with other equipment, etc.), a positive reward is given; otherwise, a negative reward is given. Through a large number of trainings, the intelligent agent model learns the optimal free position judgment strategy, and an intelligent agent model trained based on reinforcement learning can be obtained. In actual application, the initial free area segmentation result and its target feature information obtained by the U-Net semantic segmentation model are input into the trained intelligent agent model, so that the intelligent agent model can judge whether the area is really suitable for installing switchgear according to the learned strategy, and obtain the target position configuration result to further improve the accuracy and reliability of the judgment.

[0093] The present invention accurately identifies the free area in the power distribution room through the U-Net semantic segmentation model. The intelligent agent model based on reinforcement learning can intelligently judge whether the free area is suitable for installing switchgear according to the free position judgment strategy, and provide scientific and reasonable installation position suggestions; through accurate free area segmentation and position decision-making, the space of the power distribution room is maximally utilized, avoiding space waste and supporting the deployment of more equipment; the intelligent agent model considers factors such as safety distance and maintenance convenience to ensure that the installation position of the switchgear meets safety specifications and reduces potential safety hazards caused by unreasonable layout; and this solution is fully automated, reducing the manual planning cost.

[0094] In one embodiment, optimizing the target space size data and the target position configuration result by using a graph neural network model based on an attention mechanism and a reinforcement learning algorithm to obtain a target analysis and judgment result includes:

[0095] Take the targets in the distribution substation as nodes, and take the connection relationships between the targets as edges to construct a layout graph structure model of the distribution substation;

[0096] Process the layout graph structure model based on the target positions and the target space dimension data to obtain node features and edge features;

[0097] Input the node features and the edge features into a graph neural network model based on the attention mechanism for weight calculation to obtain node attention weights and edge attention weights;

[0098] With power specifications as the constraint conditions and maximizing space utilization rate as the optimization objective, optimize the target position configuration result, the node attention weights, and the edge attention weights through a reinforcement learning algorithm to obtain a layout recommendation plan as the output of the target analysis and judgment result.

[0099] Specifically, in traditional layout analysis, by regarding switch rooms and pipe galleries as nodes in a graph and their connection relationships as edges, a graph structure describing the layout of the distribution substation is constructed. Then, by using some algorithms in graph theory, such as the shortest path algorithm and connectivity analysis, the rationality of the layout can be initially analyzed. For example, if it is necessary to judge whether the wiring from a certain switch room to the pipe gallery is reasonable, it can be evaluated by calculating the corresponding shortest path in the graph. Topological analysis focuses on the relative position relationships between nodes and edges to judge whether the layout conforms to the basic spatial logic. However, the node abstraction cannot reflect the actual size and shape of the equipment, which may lead to a mismatch between the layout analysis result and the actual situation. Based on this, the present invention uses a graph neural network model GAT (Graph Attention Networks) based on the attention mechanism and a reinforcement learning algorithm to optimize the layout analysis result:

[0100] First, take the targets in the distribution substation as nodes, and take the connection relationships between the targets as edges, and use an adjacency matrix to construct a layout graph structure model of the distribution substation; then, based on the target positions and the target space dimension data, by processing the layout graph structure model, construct the features of each node and edge, that is, node features (such as the position and type of the node) and edge features (the length, connection method, etc. of the edge).

[0101] Next, input the obtained node features and edge features into the GAT model, so that the attention mechanism in the GAT model calculates the attention weights of each node and edge according to these features, that is, node attention weights (the correlation between nodes can be calculated using the multi-head attention mechanism as its weight) and edge attention weights (the importance of the edge can be calculated using a weighted combination of edge features and node features). These weights reflect the importance degrees of different nodes and edges in the layout.

[0102] Finally, based on these node attention weights and edge attention weights, with the power specification as the constraint condition and the maximization of space utilization rate as the optimization goal, the target location configuration result is optimized by combining the reinforcement learning algorithm, and a layout recommendation scheme is obtained as the output of the target analysis and judgment result. That is to say, based on the GAT evaluation, the reinforcement learning agent accurately captures the connection relationship and spatial position relationship between the weights by learning these weights, and tries different layout adjustment schemes, so as to output the layout rationality evaluation result. For example, the agent can try to move the position of a switch room or change the direction of the pipe gallery. After each adjustment, it checks whether the layout complies with the power specification according to the power specification, and calculates the space utilization rate as a reward signal to feedback to the agent, so that the agent continuously adjusts its strategy according to the reward signal. After multiple iterations, a layout recommendation that meets the power specification and has a high space utilization rate is finally obtained as the target analysis and judgment result. Or, in an actual distribution room layout, it is found through the GAT model evaluation that the connection between the current pipe gallery and some switch rooms is not reasonable and the space utilization rate is low. Then, by controlling the agent to try to adjust the direction of the pipe gallery and the position of the switch room multiple times, a new layout scheme is finally obtained as the target analysis and judgment result, which not only meets the requirements of the power specification, but also effectively improves the space utilization rate.

[0103] Through the graph neural network and the attention mechanism, the present invention accurately captures the importance of nodes and edges, making the optimization result comply with the power specification and ensuring safety and functionality; through the reinforcement learning algorithm, it learns the optimization strategy from historical data to provide a scientific layout recommendation scheme, which can be combined with the graph neural network model to further optimize the target space size data and the target location configuration result, and obtain a more accurate and feasible target analysis and judgment result; by optimizing the target position and connection relationship, the space utilization rate of the distribution room is maximized.

[0104] In one embodiment, step S4 includes: fusing the target location, target features, and target analysis and judgment results (i.e., the spatial dimensions, positions, specifications, models, and layout conditions of the distribution room and equipment, etc.) through a cross-attention mechanism, and inputting them into a deep learning model based on a multi-layer perceptron for processing, with the output being a recommended result on whether a switch cabinet can be installed. When training this deep learning model, a large amount of historical data is used, including the distribution room information and installation decisions of the already installed switch cabinets, enabling the model to learn the relationships between installation decisions and various factors. If the deep learning model predicts that a switch cabinet can be installed, detailed installation suggestions are output, including information such as the recommended installation location and the required installation space; if it cannot be installed, specific reasons are given, such as insufficient space or unreasonable layout, etc.; finally, the generated configuration strategy is sent to the distribution room control system as the remote survey and analysis result, so that it can adjust the position and angle of the switch cabinet by controlling automation devices (such as robotic arms and guide rails).

[0105] In the embodiment of the present application, aiming at the problems of how to improve the efficiency of survey and analysis of the distribution room and the efficiency and quality of the existing switch cabinet configuration method, a remote survey and analysis method for the distribution room is designed. It realizes intelligent monitoring and management of the distribution room by integrating video monitoring, lidar data, and deep learning algorithms; uses a convolutional neural network model to perform target detection and feature extraction on video monitoring data, and can accurately locate the equipment and its features in the distribution room; combines video monitoring data and lidar data, and uses deep learning algorithms and reinforcement learning algorithms for spatial analysis and layout judgment to improve space utilization; processes the fused data through a deep learning model of a multi-layer perceptron, and can formulate an optimal configuration strategy for key equipment such as switch cabinets; the entire solution realizes the automation and intelligence from data collection, processing to analysis and judgment, and configuration decision-making, greatly improving work efficiency; by fusing video monitoring data and lidar data, the advantages of different data sources are fully utilized, improving the accuracy and reliability of analysis; it can adapt to distribution rooms of different scales and types, and has strong versatility and adaptability.

[0106] It should be noted that although the steps in the above flow chart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders.

[0107] In another embodiment, as Figure 2 shown, the second aspect of the present invention provides a remote survey and analysis system for a distribution room, including:

[0108] A data acquisition module 10, configured to acquire video monitoring data and lidar data of the distribution room;

[0109] A target determination module 20, configured to perform target detection and feature extraction on the video surveillance data through a convolutional neural network model to obtain a target position and target features;

[0110] A layout analysis module 30, configured to perform spatial analysis and layout judgment on the switchgear room based on the video surveillance data and the lidar data by using a deep learning algorithm and a reinforcement learning algorithm to obtain a target analysis and judgment result;

[0111] A result output module 40, configured to fuse the target position, the target features, and the target analysis and judgment result, and input the fused result into a deep learning model based on a multi-layer perceptron for processing, so as to obtain an optimal configuration strategy for switchgear as a remote survey and analysis result for output.

[0112] It should be noted that each module in the above remote survey and analysis system for a switchgear room can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in a processor in a computer device in a hardware form or be independent of the processor, or can be stored in a memory in the computer device in a software form, so that the processor can call and execute the operations corresponding to the above modules. For the specific limitations of a remote survey and analysis system for a switchgear room, refer to the limitations of a remote survey and analysis method for a switchgear room in the above text. The two have the same functions and effects and will not be elaborated here.

[0113] The third aspect of the present invention provides an electronic device, which includes:

[0114] A processor, a memory, and a bus;

[0115] The bus is used to connect the processor and the memory;

[0116] The memory is used to store operation instructions;

[0117] The processor is configured to execute, by calling the operation instructions, executable instructions to make the processor execute the operations corresponding to a remote survey and analysis method for a switchgear room as shown in the first aspect of this application.

[0118] In an optional embodiment, an electronic device is provided, as Figure 3 shown, Figure 3 The electronic device 5000 shown includes a processor 5001 and a memory 5003. Among them, the processor 5001 and the memory 5003 are connected, such as through a bus 5002. Optionally, the electronic device 5000 may further include a transceiver 5004. It should be noted that in practical applications, the transceiver 5004 is not limited to one, and the structure of the electronic device 5000 does not constitute a limitation to the embodiments of this application.

[0119] The processor 5001 can be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logical blocks, modules, and circuits described in connection with the disclosure of this application. The processor 5001 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0120] The bus 5002 can include a path for transmitting information between the above components. The bus 5002 can be a PCI bus or an EISA bus, etc. The bus 5002 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0121] The memory 5003 can be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or it can also be an EEPROM, a CD-ROM or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0122] The memory 5003 is used to store the application program code for executing the solution of this application, and is controlled by the processor 5001 to execute. The processor 5001 is used to execute the application program code stored in the memory 5003 to implement the content shown in any of the foregoing method embodiments.

[0123] Among them, the electronic device includes but is not limited to: mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc.

[0124] The fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements a remote survey and analysis method for a power distribution room shown in the first aspect of this application.

[0125] Another embodiment of this application provides a computer-readable storage medium, on which a computer program is stored, and when it runs on a computer, it enables the computer to execute the corresponding content in the foregoing method embodiments.

[0126] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

[0127] In summary, the present invention relates to the field of information technology, and discloses a remote survey and analysis method, system, device and medium for a distribution substation. The method includes obtaining video surveillance data and lidar data of the distribution substation; performing object detection and feature extraction on the video surveillance data through a convolutional neural network model to obtain the target position and target features; based on the video surveillance data and the lidar data, using deep learning algorithms and reinforcement learning algorithms to perform spatial analysis and layout judgment on the distribution substation to obtain a target analysis and judgment result; fusing the target position, the target features and the target analysis and judgment result, and inputting the fused result into a deep learning model based on a multi-layer perceptron for processing to obtain an optimal switchgear configuration strategy as the output of the remote survey and analysis result; by integrating multiple technologies and algorithms to perform remote survey and analysis on the distribution substation, the efficiency and quality of switchgear configuration are effectively improved.

[0128] Each embodiment in this specification is described in a progressive manner. For parts that are the same or similar in each embodiment, they can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. It should be noted that the above technical features of the embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the above technical features in the embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0129] The above embodiments only represent several preferred implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art in this technical field, without departing from the technical principle of the present invention, several improvements and substitutions can still be made, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the protection scope of the claims.

Claims

1. A remote survey and analysis method for a power distribution room, characterized in that: include: Obtain video surveillance data and lidar data from the power distribution room; Performing target detection and feature extraction on the video surveillance data through a convolutional neural network model to obtain target positions and target features; Based on the video surveillance data and the laser radar data, a deep learning algorithm and a reinforcement learning algorithm are used to perform spatial analysis and layout judgment on the power distribution room to obtain a target analysis and judgment result; The target position, the target features and the target analysis and judgment results are fused and input into a deep learning model based on a multi-layer perceptron for processing, so as to obtain the optimal configuration strategy of the switch cabinet as the remote survey and analysis result output.

2. A remote survey and analysis method for a power distribution room according to claim 1, characterized in that: The step of obtaining video surveillance data of the power distribution room includes: Collect video streams of various areas in the power distribution room in real time, and decompose the video streams into a number of continuous image frames at a fixed frame rate, so that each image frame clearly reflects the real-time status in the power distribution room; Each of the image frames is normalized to eliminate the differences between the image frames, and a generative adversarial network model is used to enhance each of the normalized image frames to generate video surveillance data of the power distribution room.

3. A remote survey and analysis method for a power distribution room according to claim 1, characterized in that: The method of performing target detection and feature extraction on the video surveillance data by using a convolutional neural network model to obtain a target position and target features includes: The Faster R-CNN model based on the convolutional neural network is trained using a pre-labeled data set of distribution room equipment and scene images, and the model parameters are adjusted through the back propagation algorithm during the training process to obtain a trained Faster R-CNN model; the data set labels the location and category information of targets including switch cabinets, pipe corridors, and switch rooms; Input the video surveillance data into the trained Faster R-CNN model for target detection, so that the trained Faster R-CNN model generates target candidate regions through the region proposal network layer, and classifies and regresses the target candidate regions using the classification layer of the trained Faster R-CNN model to obtain the location and category of the target in the video surveillance data; Based on the location and category of the target in the video surveillance data, feature extraction is performed on the video surveillance data through the convolution layer and pooling layer of the trained Faster R-CNN model to obtain target features.

4. The remote survey and analysis method for a power distribution room according to claim 1, characterized in that: Based on the video surveillance data and the laser radar data, a deep learning algorithm and a reinforcement learning algorithm are used to perform spatial analysis and layout judgment on the power distribution room to obtain target analysis and judgment results, including: Based on the target position and the target features, the MonoDETR model is used to fuse the video surveillance data and the lidar data to perform spatial analysis on the power distribution room and obtain target space size data; Inputting the video surveillance data into a semantic segmentation model and an agent model based on reinforcement learning to detect vacant positions, so as to make a layout judgment on the power distribution room and obtain a target position configuration result; A graph neural network model based on an attention mechanism and a reinforcement learning algorithm are used to optimize the target space size data and the target position configuration results to obtain a target analysis and judgment result.

5. A remote survey and analysis method for a power distribution room according to claim 4, characterized in that: Based on the target position and the target features, the MonoDETR model is used to fuse the video surveillance data and the laser radar data to perform spatial analysis on the power distribution room to obtain target space size data, including: Based on the target position and the target features, the video surveillance data is encoded into a feature vector by the encoder of the MonoDETR model, and the depth information of the target in the video surveillance data is predicted according to the feature vector by the decoder of the MonoDETR model to generate a depth map; Preprocessing the laser radar data so that the preprocessed laser radar data and the video surveillance data are in the same coordinate system, and matching and fusing the depth information in the preprocessed laser radar data with the depth map to obtain optimized depth information; The optimized depth information is combined with camera parameters and pixel coordinates to quantify the target space size data of the power distribution room.

6. A remote survey and analysis method for a power distribution room according to claim 4, characterized in that: The video surveillance data is input into a semantic segmentation model and an agent model based on reinforcement learning to detect vacant positions, so as to make a layout judgment on the power distribution room and obtain a target position configuration result, including: The video surveillance data is input into a pre-trained U-Net semantic segmentation model for processing to obtain an idle area segmentation result; The idle area segmentation result is input into the intelligent agent model trained based on reinforcement learning to perform vacant position detection, so that the intelligent agent model determines whether the idle area segmentation result is suitable for installing the switch cabinet according to the trained vacant position judgment strategy, and obtains the target position configuration result.

7. A remote survey and analysis method for a power distribution room according to claim 4, characterized in that: The graph neural network model based on the attention mechanism and the reinforcement learning algorithm are used to optimize the target space size data and the target position configuration result to obtain the target analysis and judgment result, including: The targets in the power distribution room are taken as nodes, and the connection relationships between the targets are taken as edges, so as to construct a layout graph structure model of the power distribution room; Processing the layout graph structure model based on the target position and the target space size data to obtain node features and edge features; Inputting the node features and the edge features into a graph neural network model based on an attention mechanism to perform weight calculation, thereby obtaining a node attention weight and an edge attention weight; Taking the power specification as the constraint condition and maximizing the space utilization as the optimization goal, the target position configuration result, the node attention weight and the edge attention weight are optimized through the reinforcement learning algorithm to obtain the layout recommendation scheme as the target analysis and judgment result output.

8. A remote survey and analysis system for a power distribution room, characterized in that: include: Data acquisition module, used to obtain video surveillance data and lidar data of the power distribution room; A target determination module is used to perform target detection and feature extraction on the video surveillance data through a convolutional neural network model to obtain a target position and target features; A layout analysis module, used to perform spatial analysis and layout judgment on the power distribution room based on the video surveillance data and the laser radar data by using a deep learning algorithm and a reinforcement learning algorithm to obtain a target analysis and judgment result; The result output module is used to fuse the target position, the target features and the target analysis and judgment results, and input them into a deep learning model based on a multi-layer perceptron for processing, so as to obtain the optimal configuration strategy of the switch cabinet as the remote survey and analysis result output.

9. An electronic device, characterized in that: The system comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the remote survey and analysis method for a power distribution room as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the remote survey and analysis method for a power distribution room as described in any one of claims 1 to 7 is implemented.

Images