Substation surrounding environment sensitive point target identification method and system based on machine learning

Abstract

The invention provides a transformer substation surrounding environment sensitive point target identification method and system based on machine learning, and relates to the technical field of electric power engineering environment monitoring, and the method comprises the steps: obtaining a high-resolution satellite image, carrying out the preprocessing of the high-resolution satellite image, and obtaining the preprocessed data; inputting the preprocessed data into a sensitive target detection deep learning model for processing, and determining basic information of the target; performing coordinate conversion on the basic information of the target, and converting the position pixel coordinate of the target in the image into the geographic coordinate of the real world; calculating the Euclidean distance between the geographic coordinate of the converted target and the geographic coordinate of the core equipment of the transformer substation; performing risk assessment based on the Euclidean distance by using the risk quantitative assessment model, and determining the risk level of the identified target; and when the risk level exceeds a preset threshold value, early warning information is automatically generated. The method can improve the recognition precision of sensitive point targets in the surrounding environment of the transformer substation.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology in power engineering, and in particular to a method and system for identifying sensitive targets in the environment surrounding substations based on machine learning. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] Substations are critical facilities in power systems, and their safe and stable operation directly affects the reliability of the power grid. However, the noise and electromagnetic fields generated during substation operation may affect sensitive areas in the surrounding environment (such as residential areas, schools, and hospitals); at the same time, sensitive targets in the surrounding environment (such as construction sites, flammable and explosive warehouses, tall trees, dense residential areas, and large bodies of water) may also pose potential risks to the substation, such as external damage, fire, and flooding. Therefore, effective monitoring of sensitive areas around substations is crucial.

[0004] Currently, monitoring of sensitive areas around substations mainly relies on manual inspections and simple video surveillance technology. Manual inspections involve periodically dispatching inspectors to conduct on-site checks around the substation, while video surveillance uses cameras to monitor real-time footage. These methods have played a certain role in long-term practice, but they have significant limitations.

[0005] First, manual inspections have limited coverage and long cycles, making 24 / 7 real-time monitoring impossible and inefficient. Identification accuracy is highly dependent on the experience of inspection personnel, making it susceptible to subjectivity and prone to missed or incorrect assessments. Second, while video surveillance provides real-time footage, it typically requires manual review, making it difficult to automatically identify sensitive targets. Furthermore, it cannot quickly detect and issue warnings for newly added or dynamically changing sensitive targets (such as new tower cranes or stacked flammable materials). In addition, current technologies lack effective integration of multi-source data such as Geographic Information Systems (GIS), historical work orders, and real-time video, resulting in fragmented information and hindering comprehensive analysis.

[0006] In recent years, machine learning, especially deep learning, has achieved remarkable success in image recognition, making automated monitoring possible. However, directly applying it to substation environmental monitoring still faces challenges: the surrounding environment of substations is complex, target scale varies greatly, and weather and lighting conditions have a significant impact. Relying solely on visual models is prone to false positives and false negatives, and there is a lack of quantitative assessment of the relative position of targets to the substation and the risk level. Therefore, existing technologies are insufficient to meet the accuracy, real-time performance, and intelligent requirements for target identification at sensitive points in substation environments. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for identifying sensitive targets in the environment surrounding a substation based on machine learning, aiming to solve at least one of the technical problems existing in the background art.

[0008] To achieve the above objectives, the present invention provides the following solution: The first aspect of this invention provides a machine learning-based method for identifying sensitive targets in the environment surrounding a substation, comprising: Acquire high-resolution satellite imagery, and preprocess the high-resolution satellite imagery to obtain preprocessed data; The preprocessed data is input into a sensitive target detection deep learning model for processing to determine the basic information of the target; coordinate transformation is performed on the basic information of the target to convert the pixel coordinates of the target in the image into real-world geographic coordinates. Calculate the Euclidean distance between the converted geographic coordinates of the target and the geographic coordinates of the core equipment of the substation; use a risk quantification assessment model to conduct a risk assessment based on the Euclidean distance to determine the risk level of the identified target; when the risk level exceeds a preset threshold, automatically generate an early warning message.

[0009] Furthermore, the preprocessing includes: filtering, histogram equalization, spatiotemporal synchronization, and spatial mapping.

[0010] Furthermore, the sensitive target detection deep learning model uses YOLOv8n as the base model, enhances the feature pyramid network part, introduces a path aggregation network structure, and realizes bidirectional feature fusion from bottom to top and from top to bottom; after the key convolutional layer of the backbone network, a convolutional block attention module is introduced, which generates attention feature maps along the channel and spatial dimensions in sequence; and the Soft-NMS algorithm is used in post-processing.

[0011] The identification of key convolutional layers requires a multi-dimensional comprehensive analysis. The specific process involves: first, based on feature visualization, observing the degree of feature abstraction in each layer using techniques such as deconvolution, and judging the contribution by combining activation heatmaps; then, determining importance through quantitative evaluation, such as freezing training layer by layer to observe performance changes, shuffling feature maps to measure impact, and calculating gradient norms; simultaneously, referring to prior network structure, analyzing typical architectures, cross-layer connections, and channel number changes; and also considering task characteristics, as the location of key layers differs under different tasks. Finally, through preliminary screening, refined evaluation, and validation stages, multiple methods are used for cross-validation to accurately identify the key convolutional layers that have a significant impact on model performance.

[0012] As a specific implementation method, during the training of the YOLOv8n model, the samples used were high-resolution images (0.5m resolution) covering a 5km radius around the substation from the "Jilin-1" satellite, constructing a labeled dataset containing 20,000 high-resolution (resolution ≥ 0.5m) images. The optimizer used was AdamW (β1 = 0.9, β2 = 0.999), with an initial learning rate of 1e-3 and weight decay of 0.01. The learning rate scheduling employed a cosine annealing strategy, with a minimum learning rate of 1e-6 and a cycle of 50 rounds. The target confidence loss weight was set to λ1 = 0.5, and the classification loss to λ2 = 0.8. The ratio of the training set to the validation set was set to 3:2.

[0013] Furthermore, the coordinate transformation includes: The ranging mode is selected based on the target's attributes, and the ranging method corresponding to the selected mode is executed; the ranging modes include dynamic target mode and general target mode; For dynamic target modes, a multi-target tracking algorithm is used to assign unique IDs and track motion trajectories, calculating motion speed and direction. The multi-target tracking algorithm typically involves several steps. First, target detection is performed, using a detector to outline potential targets in video frames and obtain initial position and category information. Next, data association is performed, matching the detected target in the current frame with existing trajectories based on target appearance features (such as color histograms) and motion information (such as position predicted by Kalman filtering) to identify the same target. New trajectories are created for newly appearing targets, while trajectories are terminated for disappearing targets. Finally, the trajectories are optimized, such as smoothing them and handling occlusion, to improve tracking accuracy and stability, achieving continuous and stable tracking of multiple targets in the video sequence.

[0014] For multi-target tracking algorithms, Kalman filter equations are used for motion prediction, followed by Hungarian algorithm for data association, then cosine distance or Mahalanobis distance is used to measure feature similarity to determine matching detection boxes, and finally moving average filtering is used for trajectory smoothing.

[0015] For the general target mode, the pixel coordinates of the bottom center point of the target in the image are used to calculate the corresponding geodetic coordinates through collinearity equations or perspective transformation.

[0016] Furthermore, the risk quantification assessment model constructs a risk assessment function by integrating type weights, distance decay factors, dynamic coefficients, meteorological coefficients, and historical coefficients, and determines the risk level based on the output value of the risk assessment function and a preset threshold.

[0017] The risk assessment function first collects relevant target tracking data as model input, such as position deviation, abnormal speed changes, and confidence fluctuations in the detection process. Next, different types of data are assigned corresponding weights according to preset rules; for example, location information with a greater impact on risk receives a higher weight. Then, the weighted data are integrated and calculated. Finally, the calculation results are compared with preset risk level thresholds to determine the risk level of the current tracking scenario, providing an accurate basis for subsequent decision-making and ensuring stable and reliable multi-target tracking.

[0018] Furthermore, the method also includes: performing multi-terminal visualization processing on the early warning information, and displaying geographic views, video linkage, and risk heat maps on a three-dimensional visualization platform based on WebGL technology.

[0019] A second aspect of the present invention provides a machine learning-based system for identifying sensitive targets in the environment surrounding a substation, comprising: The data acquisition and processing module is used to acquire high-resolution satellite images, preprocess the high-resolution satellite images, and obtain preprocessed data. The spatial positioning module is used to input the preprocessed data into a sensitive target detection deep learning model for processing to determine the basic information of the target; and to perform coordinate transformation on the basic information of the target, converting the pixel coordinates of the target's position in the image into real-world geographic coordinates. The early warning module is used to calculate the Euclidean distance between the converted geographic coordinates of the target and the geographic coordinates of the core equipment of the substation; to perform risk assessment based on the Euclidean distance using a risk quantification assessment model to determine the risk level of the identified target; and to automatically generate early warning information when the risk level exceeds a preset threshold.

[0020] A third aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the machine learning-based method for identifying sensitive targets in the environment surrounding a substation as described in the first aspect of the present invention.

[0021] A fourth aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the machine learning-based method for identifying sensitive targets in the environment surrounding a substation as described in the first aspect of the present invention.

[0022] A fifth aspect of the present invention provides a computer program product comprising software code, wherein the program in the software code performs the steps of the machine learning-based method for identifying sensitive targets in the environment surrounding a substation as described in the first aspect of the present invention.

[0023] Compared with existing technologies, the present invention provides a machine learning-based method and system for identifying sensitive targets in the environment surrounding substations, which has the following advantages: (1) To address the limitations of existing monitoring methods in terms of coverage and real-time monitoring, this invention acquires and preprocesses high-resolution satellite images, and combines them with a deep learning model for sensitive target detection to achieve all-weather automatic monitoring of the environment surrounding substations. By leveraging the wide coverage of satellite images, the shortcomings of manual inspections are compensated for, thereby achieving a highly efficient and wide-coverage monitoring effect.

[0024] (2) To address the problem that recognition accuracy relies on human experience and is prone to missed or false positives, this invention uses YOLOv8n as the basic model and introduces a path aggregation network and a convolutional block attention module to enhance the model's ability to detect small-scale targets and complex backgrounds. By automatically recognizing targets through a deep learning model, the influence of subjectivity is reduced, thereby improving recognition accuracy and reliability.

[0025] (3) To address the problem of fragmented multi-source information and the inability to comprehensively analyze it, this invention converts the target location into geographic coordinates through coordinate transformation and integrates multiple factors such as type weight and distance decay factor using a risk quantification assessment model for risk assessment. This achieves multi-source data fusion and comprehensive analysis, thereby improving the accuracy and intelligence level of risk assessment.

[0026] (4) To address the lack of quantitative risk assessment in existing technologies, this invention constructs a quantitative risk assessment model, calculates risk levels based on Euclidean distance and multiple factors, and achieves quantitative assessment of target risks. This technical feature enables the system to issue automatic warnings, thereby realizing the transformation from "post-event emergency response" to "pre-event warning".

[0027] (5) This invention performs multi-terminal visualization processing on early warning information, and displays geographic views, video linkage, and risk heat maps based on WebGL technology. It provides an intuitive visualization interface, which facilitates quick location and decision-making by operation and maintenance personnel, thereby enhancing the availability and decision support capabilities of the system. Attached Figure Description

[0028] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0029] Figure 1 This is a flowchart of a machine learning-based method for identifying sensitive targets in the environment surrounding a substation, as provided in Embodiment 1 of the present invention. Figure 2 This is a diagram of the architecture of a deep learning model for sensitive target detection provided in Embodiment 1 of the present invention; Figure 3This is a diagram of the risk quantification assessment model architecture provided in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of a machine learning-based target identification system for sensitive points around a substation, provided in Embodiment 2 of the present invention. Figure 5 This is a schematic diagram of the environment surrounding the 110kV Bawang Substation under a high-resolution image, provided by the machine learning-based substation surrounding environment sensitive point target recognition system in Embodiment 2 of the present invention. Figure 6 This is a schematic diagram of the machine learning-based substation surrounding environment sensitive point target recognition system automatically identifying environmental sensitive points of buildings around the substation, as provided in Embodiment 2 of the present invention. Detailed Implementation

[0030] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0033] All data acquisition in this embodiment is carried out in accordance with laws and regulations and with user consent, and the data is used legally.

[0034] Example 1 like Figure 1 This embodiment provides a machine learning-based method for identifying sensitive targets in the environment surrounding a substation, including: Acquire high-resolution satellite imagery, and preprocess the high-resolution satellite imagery to obtain preprocessed data; The preprocessed data is input into a sensitive target detection deep learning model for processing to determine the basic information of the target; coordinate transformation is performed on the basic information of the target to convert the pixel coordinates of the target in the image into real-world geographic coordinates. Calculate the Euclidean distance between the converted geographic coordinates of the target and the geographic coordinates of the core equipment of the substation; use a risk quantification assessment model to conduct a risk assessment based on the Euclidean distance to determine the risk level of the identified target; when the risk level exceeds a preset threshold, automatically generate an early warning message.

[0035] Specifically, the preprocessing includes: filtering, histogram equalization, spatiotemporal synchronization, and spatial mapping.

[0036] Specifically, such as Figure 2 The sensitive target detection deep learning model uses YOLOv8n as the base model, enhances the feature pyramid network part, introduces a path aggregation network structure, and realizes bidirectional feature fusion from bottom to top and from top to bottom. A convolutional block attention module is introduced after the key convolutional layer of the backbone network. The convolutional block attention module generates attention feature maps along the channel and spatial dimensions in sequence. The Soft-NMS algorithm is used in post-processing.

[0037] Specifically, the coordinate transformation includes: The ranging mode is selected based on the target's attributes, and the ranging method corresponding to the selected mode is executed; the ranging modes include dynamic target mode and general target mode; For dynamic target mode, a multi-target tracking algorithm is used to assign a unique ID and track the motion trajectory to calculate the motion speed and direction; For the general target mode, the pixel coordinates of the bottom center point of the target in the image are used to calculate the corresponding geodetic coordinates through collinearity equations or perspective transformation.

[0038] Specifically, such as Figure 3 The risk quantification assessment model constructs a risk assessment function by integrating type weights, distance decay factors, dynamic coefficients, meteorological coefficients, and historical coefficients, and determines the risk level based on the output value of the risk assessment function and a preset threshold.

[0039] As a specific implementation method, the risk quantification assessment model is constructed from type weights, distance decay factors, dynamic coefficients, meteorological coefficients, and historical coefficients. The final risk value divides the target into three risk levels according to a preset threshold. The distance decay factor adopts an exponential decay function.

[0040] Regarding the risk level setting, a large amount of prior data was first obtained based on preset experiments, and the data status in the experimental scenario was analyzed. Two thresholds were reasonably set. When the final risk value is lower than the first lower threshold, the target is classified as low-risk, meaning its tracking status is stable and there are few interfering factors, allowing for a reduction in computational resource allocation. If the risk value is between the two thresholds, it is classified as medium-risk, requiring continuous close monitoring. When the risk value is higher than the higher threshold, it is classified as high-risk, indicating that the target may exhibit abnormalities such as severe occlusion or rapid movement. In this case, an enhanced tracking mechanism should be activated immediately to ensure the target is not lost. The exponent base of the exponential decay function is set to 0.8, and the decay rate is set to 0.12.

[0041] Specifically, the method also includes: performing multi-terminal visualization processing on the early warning information, and displaying geographic views, video linkage, and risk heat maps on a three-dimensional visualization platform based on WebGL technology.

[0042] The geographic view is used to dynamically mark all identified sensitive points on the 3D real-scene model or 2D map of the substation with red, orange and yellow icons, and clicking the icon can view details; the video linkage is used to click the icon on the map or the alarm list; the risk heat map is used to generate a risk heat map of the surrounding area based on all identified events within a period of time.

[0043] In one specific embodiment, the method provided by the present invention includes the following steps: Step S1: Data acquisition and preprocessing.

[0044] S1.1: Data Acquisition: The system accesses high-resolution (0.5-meter level) remote sensing images from commercial satellites such as "Jilin-1," updated weekly. Simultaneously, it accesses video streams from four 2-megapixel high-definition PTZ cameras deployed around the substation perimeter, enabling 24 / 7 uninterrupted monitoring.

[0045] S1.2: Preprocessing: Filtering: Gaussian filtering is applied to the video stream to effectively suppress noise interference at night.

[0046] Histogram equalization: Enhances the contrast of images taken in overcast or rainy weather or at dusk, making targets clearer.

[0047] Spatiotemporal synchronization and spatial mapping: Using satellite imagery as a base map, a mapping relationship between each target area and its real geographic coordinates is established through a GIS platform to prepare for subsequent coordinate transformation.

[0048] S2: Construction and training of a deep learning object detection model based on multi-scale feature fusion.

[0049] S2.1: Model Architecture Selection and Improvement: Using YOLOv8n as the base model for lightweight deployment, while ensuring speed, we enhance its Feature Pyramid Network (FPN) part and introduce the Path Aggregation Network (PANet) structure to achieve bidirectional feature fusion from bottom to top and from top to bottom, which significantly improves the detection capability of small-scale targets (such as distant machinery and personnel).

[0050] A Convolutional Block Attention (CBAM) module is introduced after the key convolutional layers of the backbone network. This module generates attention feature maps sequentially along the channel and spatial dimensions, allowing the model to focus more on salient regions in the image (such as construction areas, smoke, etc.) and suppress interference from complex backgrounds.

[0051] For targets that are partially occluded, the Soft-NMS (Non-Maximum Suppression) algorithm is used instead of the traditional NMS. By reducing the confidence of overlapping boxes instead of deleting them directly, more likely detection results are retained, thus improving the recall rate.

[0052] S2.2: Refined Dataset Construction and Enhancement: Data annotation: Tools such as LabelImg were used to perform detailed annotations on the collected substation scene images. Not only were the bounding boxes and categories of the targets annotated, but key points of certain targets (such as crane booms) were also annotated to determine their orientation and activity status.

[0053] Data augmentation: Mosaic augmentation is employed, stitching four training images together to increase the number of targets in a single image and improve the model's generalization ability. Simultaneously, different weather and lighting conditions are simulated, and algorithms such as randomly adjusting brightness, contrast, and saturation, adding Gaussian noise, and simulating rain and snow stripes are applied to greatly expand the diversity of the training set, making the model highly adaptable to changes in the outdoor environment.

[0054] S2.3: Model Training and Optimization A transfer learning strategy was employed, using weights pre-trained on the MS COCO dataset as initial values. The AdamW optimizer was used, whose weight decay strategy helps prevent overfitting. The initial learning rate was set to 1e-3, and a cosine annealing strategy was used to dynamically adjust the learning rate.

[0055] Define the weighted loss function: Total Loss = CIoU Loss + λ1 * Target Confidence Loss + λ2 * Classification Loss. To improve positioning accuracy, CIoU Loss is used instead of the traditional IoU Loss, as it considers overlap area, center point distance, and aspect ratio.

[0056] S3: Multimodal information perception and precise geographic positioning of targets.

[0057] S3.1: Target perception and attribute extraction: The preprocessed real-time images are input into the trained model to obtain the bounding box, class label, and confidence score for each target. For dynamic targets (such as vehicles and people), multi-target tracking algorithms such as ByteTrack are used to assign them unique IDs, track their motion trajectories, calculate their speed and direction, and determine whether they exhibit threatening behaviors such as "loitering" or "approaching".

[0058] S3.2: Monocular visual ranging and geographic coordinate calculation: For more general scenarios, the pixel coordinates (u, v) of the target's bottom center point in the image (assuming this point touches the ground) are used to calculate its corresponding geodetic coordinates (B, L) through collinearity equations or perspective transformation. The calculation process involves solving for the rotation matrix R and the translation vector T, ultimately outputting the target's latitude and longitude.

[0059] S4: Construct and assess dynamic risk maps based on GIS for comprehensive analysis.

[0060] S4.1: Spatial Relationship and Situation Calculation: In a GIS engine (such as ArcGIS Engine or SuperMap), the calculated target latitude and longitude are used to generate a spatial point layer. Multi-level buffer rings are generated based on the substation boundary (e.g., 50-meter core restricted area, 300-meter warning area, 500-meter monitoring area).

[0061] Perform spatial query and analysis: calculate the minimum electrical distance (not the straight-line distance) between the target and the substation; determine whether the target is located within the buffer zone; calculate the vertical distance and clearance distance between the target and the high-voltage line (requires importing DEM digital elevation model data).

[0062] S4.2: Multi-factor dynamic risk assessment model: Construct a risk scoring function, including: Type weight: Assign a basic risk value to different targets (e.g., open flame = 10, tower crane = 8, construction machinery = 6, tall trees = 5, ordinary vehicles = 3).

[0063] Distance decay factor: Using an exponential decay function, the closer the distance, the larger the factor and the higher the risk.

[0064] Dynamic coefficients: The static target coefficient is 1, the target coefficient for moving to the substation is 1.5, and the target coefficient for high-speed movement is further increased.

[0065] Meteorological factors: In windy weather (wind speed > 10 m / s), the risk factor of trees and temporary buildings increases; in foggy weather (visibility < 100 m), the overall monitoring uncertainty of all targets increases.

[0066] Historical coefficient: If the location has experienced external damage or warnings within the past year, the coefficient increases.

[0067] The final risk value is obtained based on the above parameters, and the target is divided into three risk levels: "red (high), orange (medium), and yellow (low)" according to the preset threshold.

[0068] S5: Intelligent early warning and visual interaction.

[0069] S5.1: Tiered Early Warning and Information Generation The system automatically triggers different levels of alerts based on the risk level. A red alert triggers an audible and visual alarm and sends an SMS to the relevant personnel; an orange alert pops up a notification on the monitoring center's large screen; and a yellow alert is only marked on the system log and map.

[0070] The structured generation of early warning information includes: time, location (latitude and longitude and text description), target category, risk level, on-site snapshot, real-time video stream link, brief description of risk assessment basis, and handling suggestions (such as "it is recommended to conduct on-site verification within 24 hours").

[0071] S5.2: Multi-dimensional Visual Cockpit A 3D visualization platform developed based on WebGL technology seamlessly integrates 2D GIS maps, video feeds, and data dashboards.

[0072] Geographic View: On the 3D reality model or 2D map of the substation, all identified sensitive points are dynamically marked with red, orange, and yellow icons. Clicking on an icon will display details.

[0073] Video linkage: Click on the icon on the map or the alarm list to achieve "one-click tracking".

[0074] Risk Heat Map: Based on all identified events over a period of time, a risk heat map of the surrounding area is generated, which intuitively displays high-risk areas and provides data support for inspection route planning.

[0075] Therefore, the beneficial effects of the present invention are as follows: Automation and high efficiency: It enables 24 / 7 unmanned automatic monitoring of the substation's surrounding environment, greatly improving inspection efficiency and freeing up manpower.

[0076] High accuracy and low false alarms: Through optimized deep learning models and multi-source information fusion (visual + GIS + meteorological), the false alarms and missed alarms caused by a single sensor are effectively reduced, resulting in high recognition accuracy.

[0077] Proactive early warning and risk quantification: It can not only identify targets, but also accurately locate them geographically and conduct risk quantification assessments, realizing the transformation from "post-event emergency response" to "pre-event early warning" and improving the safety protection level of substations.

[0078] Decision support and visualization: Provides an intuitive visualization interface and rich contextual information, offering powerful data support for operations and maintenance personnel's decision-making, facilitating rapid response and handling.

[0079] Example 2 like Figure 4 As shown, this embodiment provides a machine learning-based system for identifying sensitive targets in the environment surrounding a substation, including: The data acquisition and processing module is used to acquire high-resolution satellite images, preprocess the high-resolution satellite images, and obtain preprocessed data. The spatial positioning module is used to input the preprocessed data into a sensitive target detection deep learning model for processing to determine the basic information of the target; and to perform coordinate transformation on the basic information of the target, converting the pixel coordinates of the target's position in the image into real-world geographic coordinates. The early warning module is used to calculate the Euclidean distance between the converted geographic coordinates of the target and the geographic coordinates of the core equipment of the substation; to perform risk assessment based on the Euclidean distance using a risk quantification assessment model to determine the risk level of the identified target; and to automatically generate early warning information when the risk level exceeds a preset threshold.

[0080] In one specific embodiment, such as Figure 5 and Figure 6 As shown, the inventors' team used the aforementioned machine learning-based substation environmental sensitive point target identification system to monitor the environment around the Bawang substation in real time, successfully identifying multiple environmental sensitive points, including construction tower cranes, temporary prefabricated houses, large vehicles, and nearby tall trees. The system accurately outputs the category, geographical coordinates, and precise distance from the core area of ​​the substation for each sensitive target. This verifies that the method provided in Embodiment 1 of this invention possesses comprehensive capabilities of high-precision identification, real-time early warning, and intuitive visualization in complex real-world scenarios, significantly improving the intelligence level and proactive defense capabilities of substation environmental safety monitoring.

[0081] Example 3 Embodiment 3 of the present invention provides an electronic device.

[0082] An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, it implements the steps in the machine learning-based method for identifying sensitive targets around a substation.

[0083] The detailed steps are the same as those of the machine learning-based target identification method for sensitive points around substations provided in Example 1, and will not be repeated here.

[0084] Example 4 Embodiment 4 of the present invention provides a computer-readable storage medium.

[0085] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the machine learning-based method for identifying sensitive targets in the environment surrounding a substation as described in Embodiment 1 of the present invention.

[0086] The detailed steps are the same as those of the machine learning-based target identification method for sensitive points around substations provided in Example 1, and will not be repeated here.

[0087] Example 5 Embodiment 5 of the present invention provides a computer program product.

[0088] A computer program product includes software code, wherein the program in the software code performs the steps of the machine learning-based method for identifying sensitive targets in the environment surrounding a substation as described in Embodiment 1 of the present invention.

[0089] The detailed steps are the same as those of the machine learning-based target identification method for sensitive points around substations provided in Example 1, and will not be repeated here.

[0090] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0091] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0095] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0096] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for identifying sensitive targets in the environment surrounding a substation based on machine learning, characterized in that, include: Acquire high-resolution satellite imagery, and preprocess the high-resolution satellite imagery to obtain preprocessed data; The preprocessed data is input into a sensitive target detection deep learning model for processing to determine the basic information of the target. The basic information of the target is transformed by coordinate transformation, converting the pixel coordinates of the target's location in the image into real-world geographic coordinates; Calculate the Euclidean distance between the converted geographic coordinates of the target and the geographic coordinates of the core equipment of the substation; The risk level of the identified target is determined by using a risk quantification assessment model based on the Euclidean distance. When the risk level exceeds a preset threshold, an early warning message is automatically generated.

2. The method as described in claim 1, characterized in that, The preprocessing includes: filtering, histogram equalization, spatiotemporal synchronization, and spatial mapping.

3. The method as described in claim 1, characterized in that, The sensitive target detection deep learning model uses YOLOv8n as the base model, enhances the feature pyramid network, introduces a path aggregation network structure to achieve bidirectional feature fusion from bottom to top and from top to bottom; a convolutional block attention module is introduced after the key convolutional layer of the backbone network, and the convolutional block attention module generates attention feature maps along the channel and spatial dimensions in sequence; and the Soft-NMS algorithm is used in post-processing.

4. The method as described in claim 1, characterized in that, The coordinate transformation includes: The ranging mode is selected based on the target's attributes, and the ranging method corresponding to the selected mode is executed; the ranging modes include dynamic target mode and general target mode; For dynamic target mode, a multi-target tracking algorithm is used to assign a unique ID and track the motion trajectory to calculate the motion speed and direction; For the general target mode, the pixel coordinates of the bottom center point of the target in the image are used to calculate the corresponding geodetic coordinates through collinearity equations or perspective transformation.

5. The method as described in claim 1, characterized in that, The risk quantification assessment model constructs a risk assessment function by integrating type weights, distance decay factors, dynamic coefficients, meteorological coefficients, and historical coefficients, and determines the risk level based on the output value of the risk assessment function and a preset threshold.

6. The method as described in claim 1, characterized in that, The method also includes: performing multi-terminal visualization processing on early warning information, and displaying geographic views, video linkage, and risk heat maps on a three-dimensional visualization platform based on WebGL technology.

7. A machine learning-based system for identifying sensitive targets in the environment surrounding a substation, characterized in that, include: The data acquisition and processing module is used to acquire high-resolution satellite images, preprocess the high-resolution satellite images, and obtain preprocessed data. The spatial positioning module is used to input the preprocessed data into the sensitive target detection deep learning model for processing to determine the basic information of the target; The basic information of the target is transformed into coordinates, converting the pixel coordinates of the target's location in the image into real-world geographic coordinates. The early warning module is used to calculate the Euclidean distance between the converted geographic coordinates of the target and the geographic coordinates of the core equipment of the substation. The risk level of the identified target is determined by using a risk quantification assessment model based on the Euclidean distance. When the risk level exceeds a preset threshold, an early warning message is automatically generated.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the steps of the machine learning-based target identification method for sensitive points in the substation surrounding environment as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the machine learning-based target identification method for sensitive points in the environment surrounding a substation as described in any one of claims 1 to 6.

10. A computer program product, comprising software code, characterized in that, The program in the software code performs the steps of the machine learning-based target identification method for sensitive points in the substation surrounding environment as described in any one of claims 1 to 6.

Images