Substation intelligent settlement inspection system and method based on three-dimensional model

Abstract

The invention relates to the technical field of substation inspection, and particularly discloses an intelligent substation settlement inspection system and method based on a three-dimensional model. The system adopts a four-layer architecture of a sensing layer, a network layer, a platform layer and an application layer, wherein the sensing layer acquires settlement data through a multi-source heterogeneous sensor network such as a static leveling instrument, a GNSS displacement monitoring station and an unmanned aerial vehicle aerial photography system; the network layer transmits data through an industrial Internet of Things gateway in a wired / wireless mode; the platform layer binds data to a three-dimensional model through a data fusion and mapping module, performs settlement trend prediction by using an LSTM neural network, and dynamically renders a settlement state through a three-dimensional visualization engine; the application layer provides a Web and mobile terminal interaction interface. The method comprises the steps of multi-source data acquisition, model binding, intelligent prediction, visual rendering and multi-level early warning closed-loop processing. According to the invention, the problems of low efficiency and data isolation of traditional settlement monitoring are solved, all-weather and high-precision intelligent monitoring and active early warning of substation settlement are realized, and the safety operation and maintenance level of a power grid is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of substation inspection technology, and in particular to a substation intelligent settlement inspection system and method based on a three-dimensional model. Background Technology

[0002] Substations are critical hubs in the power system. Settlement and deformation of their foundations, equipment bases, and structures can lead to twisting, deformation, or even fracture of equipment foundations, causing structural deformation, clamp breakage, cable breakage, and even equipment collapse. Heavy equipment such as transformers, in particular, has extremely high requirements for foundation flatness; even minor deformations can increase internal mechanical stress, affecting insulation performance and normal operation, directly impacting the safe and stable operation of the power grid. Therefore, systematic monitoring and inspection of substation settlement and deformation is an indispensable part of ensuring the safe and stable operation of the power grid.

[0003] Currently, traditional settlement monitoring mainly relies on manual measurement using equipment such as levels and total stations at regular intervals. This method has inherent drawbacks such as low efficiency, poor real-time performance, isolated data, and difficulty in capturing continuous deformation trends. Although some existing automated monitoring systems can automatically collect data, they usually rely on single-type sensors and lack the ability to deeply fuse and analyze multi-source heterogeneous data. At the same time, their data display is mostly based on two-dimensional charts, which cannot intuitively and accurately present the distribution and dynamic evolution of settlement in three-dimensional space, making it difficult to effectively support accurate fault diagnosis and maintenance decisions.

[0004] To address these issues, a substation intelligent settlement inspection system and method based on a 3D model are provided. Summary of the Invention

[0005] This invention provides a three-dimensional model-based intelligent substation settlement inspection system and method, which aims to achieve all-weather, high-precision, intelligent monitoring and multi-level early warning of substation settlement by constructing a multi-source heterogeneous sensor collaborative sensing network, a high-precision three-dimensional digital twin model, and integrating multi-modal data for intelligent analysis and judgment.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a substation intelligent settlement inspection system based on a three-dimensional model, comprising a perception layer, a network layer, a platform layer and an application layer; The perception layer consists of a multi-source heterogeneous sensor network deployed within the substation, used to collect settlement monitoring data. The sensor network includes, but is not limited to: a hydrostatic level (deployed at key feature points of the main transformer, GIS equipment, and infrastructure to form a closed measurement loop) for relative settlement monitoring; a GNSS displacement monitoring station (composed of a reference station and a monitoring station, employing real-time dynamic carrier phase differential technology) for absolute displacement monitoring; a drone aerial photography system for periodically collecting high-precision three-dimensional point cloud and real-scene image data across the entire area; a fixed monitoring camera for dual-spectrum video monitoring using visible light and thermal imaging; and an inspection robot with autonomous navigation capabilities for mobile supplementary monitoring. The network layer is used to transmit the monitoring data of the perception layer; the data is aggregated using an industrial IoT gateway, and the encrypted data is transmitted to the platform layer via wired (such as industrial Ethernet) or wireless (such as wireless private network, 5G) communication methods. The platform layer serves as the core processing unit of the system, used for the fusion, analysis, visualization, and early warning processing of monitoring data. The application layer provides users with access interfaces in the form of web browsers and mobile apps to realize functions such as visual query of monitoring data, alarm information reception, inspection task management and report generation.

[0007] As a further description of the above technical solution: The platform layer includes: ① The data fusion and mapping module is the key to solving the problem of accurate correlation of multi-source data, and it adopts a dual correlation and cross-validation mechanism; a. Predefined mapping table association: For sensors with fixed locations (such as hydrostatic level), a mapping table is established between sensor IDs and unique codes of BIM model components to achieve direct and accurate association between data and model components; b. Dynamic spatial coordinate association: For data that provides absolute geographic coordinates (such as GNSS and UAV data), the spatial nearest neighbor query algorithm (specifically by calculating the Euclidean distance between the sensor coordinates and the vertices of the triangulation network of the real scene model) is used to dynamically bind it to the nearest vertex or virtual measurement point. c. Cross-validation and calibration: The two types of correlation results are compared and validated in real time. A calibration alarm is automatically triggered when discrepancies occur, ensuring extremely high reliability and accuracy of the data mapping. This mechanism effectively solves the challenges of depth and reliability in multi-source heterogeneous data fusion.

[0008] ② The intelligent settlement analysis module is the core of intelligent early warning. It integrates a prediction unit based on LSTM (Long Short-Term Memory) neural network. This unit takes the historical settlement time series as input and outputs the settlement prediction value and the probability of exceeding the limit for a period of time in the future through deep learning, realizing the leap from passive response to active prediction.

[0009] ③ 3D Visualization Engine: This module is responsible for achieving immersive visualization. It adopts instantiated rendering and LOD (Level of Detail) technology, and uses algorithms such as Quadric Mesh Simplification to lightweightly integrate high-precision BIM models with real-world models. Based on monitoring data, the engine renders the settlement status in real time and dynamically in 3D space through color mapping (dynamically generating colors according to alarm thresholds) and vertex displacement (dynamically adjusting the position of model mesh vertices through vertex shaders), making the settlement situation clear at a glance.

[0010] ④ The early warning and response management module is the guarantee for achieving a closed-loop operation and maintenance system, and it adopts a multi-level alarm mechanism. This mechanism is configured as follows: Level 1 alarm: triggered when the cumulative settlement exceeds the design allowable value and the duration exceeds the limit (e.g., 2 hours); Level 2 alarm: triggered when the settlement rate exceeds the set threshold (e.g., 0.2 mm / day) and exceeds the limit for multiple consecutive sampling cycles (e.g., 3 cycles); Level 3 warning: triggered when the probability of future settlement exceeding the limit predicted by the LSTM neural network is greater than a set threshold (e.g., 80%).

[0011] Secondly, the present invention provides a method for intelligent settlement inspection of substations based on a three-dimensional model, comprising the following steps: S1: Settlement data is collected through a multi-source sensor network; S2: Dynamically bind data to the 3D model through a dual association and cross-validation mechanism; S3: Predicting subsidence trends based on LSTM neural networks; S4: Perform dynamic visualization rendering in the merged 3D model; S5: Triggers multi-level alarms and initiates closed-loop handling procedures.

[0012] The present invention has the following beneficial effects: The intelligent substation settlement inspection system and method based on a three-dimensional model described in this invention solves the technical problem of high-precision binding between multi-source heterogeneous data and three-dimensional models through an original dual association and cross-verification mechanism, realizing the leap from "data silos" to "deep integration" and laying a solid foundation for accurate analysis. The substation intelligent settlement inspection system and method based on a three-dimensional model is based on a lightweight fusion model and dynamic visualization rendering technology, which realizes the real-time, intuitive and immersive presentation of settlement status in three-dimensional space, greatly improving the status perception ability and decision-making efficiency of operation and maintenance personnel. The substation intelligent settlement inspection system and method based on a three-dimensional model, by introducing an LSTM neural network prediction model and a multi-level alarm threshold mechanism, realizes early identification and probabilistic early warning of settlement trends, transforming the passive mode of "post-event remediation" into the proactive mode of "pre-event prevention", significantly reducing safety risks. The substation intelligent settlement inspection system and method based on three-dimensional model establishes a fully automated closed-loop management mechanism from automatic early warning, intelligent review to disposal verification, which greatly reduces manual intervention, improves operation and maintenance response speed and disposal efficiency, and ensures process traceability and continuous optimization. The aforementioned intelligent substation settlement inspection system and method based on a 3D model adopts a modular design and standardized interface, supports flexible access to various sensors and models, and has good adaptability and scalability, providing strong technical support for the construction of smart substations. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall architecture of the intelligent substation settlement inspection system based on a three-dimensional model as described in this invention. Figure 2 This is a flowchart of the data fusion and dynamic mapping module mechanism of the present invention; Figure 3 This is a schematic diagram of the structure of the LSTM neural network prediction unit of the present invention; Figure 4 This is a flowchart of the three-dimensional dynamic visualization rendering process of the present invention; Figure 5 This is a flowchart illustrating the closed-loop management process for early warning and response in this invention. Figure 6 The overall flowchart for inventing a substation intelligent settlement inspection method based on a 3D model. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Example 1: System deployment and data fusion processing, such as Figure 1 , Figure 2 As shown This system was deployed at a 500kV substation located in a coastal soft soil area, where uneven settlement had historically caused the GIS equipment busbar to crack.

[0016] 1. System hardware deployment and configuration The multi-source heterogeneous sensor network of the perception layer is structured as follows: Hydrostatic leveling instrument: A liquid level interconnection sensor system with an accuracy of ±0.1mm is selected. Five measuring points are set up at the four corners and the center of each main transformer foundation (2 units in total), with the distance between adjacent measuring points controlled within 15 meters; three measuring points are set up at key locations in each bay of the 220kV GIS equipment foundation (4 sets), for a total of 12 measuring points; two measuring points are set up at each support foundation of the 500kV structure foundation (10 spans), for a total of 20 measuring points; all measuring points are connected to the station edge computing gateway via RS485 bus, and the sampling frequency is set to 1 time / minute. GNSS displacement monitoring station: RTK receivers supporting both BeiDou-3 and GPS systems are selected, with accuracy down to the millimeter level; the base station is installed on bedrock 1.5 kilometers from the substation, using concrete observation piers for forced centering; monitoring stations (6 in total) are installed on the roof of the main control building (2 units), the roof of the 220kV relay room (2 units), and the ground-based forced centering observation piers in the station area (2 units), with a sampling frequency set to 1Hz, and data is transmitted in real time with encryption via a built-in 4G / 5G communication module; Drone aerial photography system: DJI Matrice 350 RTK drone is selected, equipped with Zenmuse P1 (45-megapixel oblique photography camera) and L2 (LiDAR) payload, to carry out high-precision data collection (resolution up to 5mm) of the entire substation area once every quarter, and to start emergency aerial photography in special circumstances (such as after heavy rain); Fixed surveillance cameras and inspection robots: Fixed cameras are deployed in key areas and support dual-spectrum monitoring of visible light and thermal imaging; inspection robots have autonomous navigation capabilities and are used for mobile supplementary monitoring.

[0017] The network layer uses industrial IoT gateways (such as Huawei AR550 series) to aggregate sensor data, which is then transmitted to the platform layer server via the on-site wired fiber optic network. Off-site data is transmitted encrypted via a 5G VPN tunnel.

[0018] The platform layer is deployed on a cloud server, using a microservice architecture and Docker containerization. The hardware configuration is: Intel Xeon Platinum processor, 128GB RAM, and NVIDIA Tesla V100 GPU, used for LSTM model training and inference.

[0019] 2. Data fusion and mapping processing The platform-level data fusion and mapping module performs the following operations: 1) Data Access and Preprocessing: Real-time data streams are received via the MQTT protocol, and data is cleaned using the following methods: a. Sliding window jump rejection: A sliding window algorithm with a window size of 10 sampling points is used to reject gross errors; b. Kalman filter smoothing: Smooths the data sequence to suppress random noise; c. Time series imputation: For missing data, the ARIMA model is used for imputation to ensure data continuity; 2) Mapping table association: The mapping relationship between the static level instrument ID and the unique code of the BIM model component (such as "main transformer 1#-foundation-B1") is pre-entered in the system; when the data arrives, the mapping table is queried to accurately associate the data with the corresponding BIM component; 3) Spatial coordinate association: For GNSS and UAV data, analyze their geographic coordinates (WGS-84 coordinate system), and calculate the Euclidean distance between the sensor coordinates and the vertices of the triangulation network of the real scene model through the spatial nearest neighbor query algorithm. Bind the data to the vertex with the smallest distance or the preset virtual measurement point (e.g., the distance threshold is set to 5cm). 4) Cross-validation: The results of mapping table association and spatial coordinate association are compared in real time; if the deviation exceeds the tolerance (e.g., horizontal deviation > 10mm or vertical deviation > 5mm), a calibration alarm is triggered, the system automatically records the log and notifies the operation and maintenance personnel to intervene.

[0020] Example 2: Intelligent analysis and visualization rendering, such as Figure 3 , Figure 4 As shown 1. Intelligent Settlement Analysis The intelligent settlement analysis module is the core processing unit of the system, integrating a prediction model based on LSTM (Long Short-Term Memory) neural networks for settlement trend prediction and multi-level alarm triggering. The detailed implementation process is as follows: 1) LSTM Neural Network Architecture and Configuration a. Network structure: A three-layer LSTM unit stacked architecture is adopted, with specific parameters as follows: Input layer: Receives historical settlement time series data, with an input dimension of 720 (i.e., sampling points per hour for 30 days).

[0021] Hidden layers: The first LSTM layer has 128 units, the second layer has 64 units, and the third layer has 32 units; the activation function is ReLU, and the dropout rate is set to 0.2 to prevent overfitting.

[0022] Output layer: Fully connected layer, outputs the predicted settlement value (unit: mm) and the probability of exceeding the limit (range 0~1) for the next 168 hours (7 days).

[0023] b. Training data preparation: Historical settlement data from the past two years will be used, covering various operating conditions (such as rainy season, freeze-thaw period, and high-temperature period). Data preprocessing includes: Normalization: Min-Max scaling is used to map the data to the [0, 1] interval.

[0024] Data augmentation: Training samples were generated by sliding a time window (window size 720, step size 24), resulting in a total of 10,000 samples.

[0025] Dataset split: 80% training set, 10% validation set, and 10% test set.

[0026] c. Model Training: TensorFlow 2.8 framework was used, with Adam as the optimizer (learning rate 0.001, decay rate 0.9). The loss function was a combination of mean squared error (MSE) and binary cross-entropy (for probability output). The training cycle was 100 epochs, with an early stopping mechanism (patience=10) to prevent overfitting. After training, the model's mean absolute error (MAE) on the test set was 0.12 mm, and the probability prediction accuracy reached 92%.

[0027] d. Real-time prediction: The model is deployed on a platform-level GPU server, with an inference frequency of once per hour. It takes the latest 30 days of data as input, outputs predictions for the next 7 days, and calculates the probability of exceeding limits (based on the cumulative distribution of predicted values ​​exceeding the design allowable value by 10 mm).

[0028] 2) Implementation of multi-level alarm mechanism The alarm mechanism is dynamically triggered based on LSTM output and real-time monitoring data. The specific parameter configuration is as follows: Level 1 Alarm: Triggered when cumulative settlement > 10mm (design allowable value) and the duration of this state > 2 hours. The system automatically records the alarm event and pushes notifications through the application layer.

[0029] Level 2 alarm: Triggered when the settlement rate exceeds 0.2 mm / day and exceeds the limit for 3 consecutive sampling cycles (1 hour per cycle). After triggering, the system starts the automatic verification process.

[0030] Level 3 warning: Triggered when the probability of settlement exceeding the limit predicted by LSTM within the next 3 days is > 80%. After the warning event is generated, the system initiates a trend analysis report for expert evaluation.

[0031] The alarm threshold supports dynamic adjustment: The system reviews historical alarm data monthly and optimizes the threshold parameters through reinforcement learning algorithms (such as adaptively adjusting the settlement rate threshold to the range of 0.15-0.25 mm / day).

[0032] 2. Implementation of 3D visualization engine The 3D visualization engine is responsible for fusing monitoring data with 3D models and achieving immersive visualization through dynamic rendering technology; technical details are as follows: 1) Lightweight Model Processing BIM Model Processing: The IFC format BIM model provided by the design institute (originally with approximately 20 million faces) is lightweighted through the following steps: Format conversion: Use the IfcOpenShell tool to convert IFC to glTF format, preserving the component hierarchy and attribute information.

[0033] Mesh simplification: The Quadric Mesh Simplification algorithm is adopted, with the number of iterations set to 5, reducing the target number of faces to below 5 million (actual output 4.8 million faces), and the error is controlled to be less than 0.1mm.

[0034] Instantiated rendering: Instantiate repetitive components (such as bolts and brackets) to reduce the number of drawing calls and improve rendering efficiency.

[0035] Reality model processing: OSGB format reality models captured by drones are converted to 3D Tiles format using ContextCapture software, supporting streaming loading. Point cloud data is compressed using LASzip, achieving a compression ratio of 1:10.

[0036] Dynamic rendering a. Color mapping mechanism: Colors are dynamically generated based on alarm thresholds: normal state (green, RGB: 0, 255, 0), warning state (yellow, RGB: 255, 255, 0), and alarm state (red, RGB: 255, 0, 0). Rendering process: The fragment shader interpolates and calculates color values ​​based on real-time monitoring data and applies them to the corresponding model components; Transparency settings: Alert state alpha=0.7, alarm state alpha=1.0; b. Vertex displacement mechanism: Displacement calculation: Based on the settlement data (unit mm), scaled proportionally (scaling factor 0.1, i.e., 1 mm settlement corresponds to 0.1 pixel displacement), the position of the mesh vertices is dynamically adjusted through the vertex shader.

[0037] Deformation simulation: For BIM model components, the displacement direction is along the normal direction; for real-world models, the displacement is calculated based on the surface normal vector of the triangular mesh; Displacement smoothing: Bilinear interpolation is used to avoid jagged edges.

[0038] c. Real-time interactive functionality: The engine supports WebGL 2.0, allowing users to perform the following operations through a browser: Sectional analysis: Dynamically generate horizontal and vertical sections to examine internal settlement conditions; Slide the timeline: replay historical subsidence animations and observe the evolution trend; Click to query: Click on a model component to bring up an information panel that displays real-time data, historical curves, and prediction results.

[0039] 3. Practical Application and Performance Testing After deployment at a 500kV substation, a 90-day application test was conducted. Intelligent analysis performance: The LSTM model has a prediction accuracy of 88% (for the next 7 days), a false alarm rate of less than 5%, and an alarm response delay of less than 3 seconds.

[0040] Performance visualization: In the Chrome browser, the rendering frame rate is stable at 60 FPS, and the model loading time is less than 10 seconds (100Mbps network).

[0041] User feedback: Maintenance personnel have reduced their average decision-making time by 50% and improved anomaly detection efficiency by 70% through the visual interface.

[0042] Example 3: Early warning and closed-loop handling process, such as Figure 5 , Figure 6 As shown 1. Early warning triggering and work order generation When the system triggers a level-two alarm (such as exceeding the settlement rate limit), the early warning and response management module will automatically execute: Generate alarm events (ID format such as "20241015-ALM-002") and record information such as time, location, and data value.

[0043] Automatically create inspection work orders and dispatch them to the mobile app of maintenance personnel via RabbitMQ message queue.

[0044] 2. UAV verification and deformation confirmation After the maintenance personnel confirm the work order, the system instructs the drone hangar to automatically take off the drone: The drone autonomously plans its flight path and performs emergency oblique photography and LiDAR scanning of the alarm area.

[0045] The returned data is processed by the platform layer to generate a realistic 3D model. The Iterative Closest Point (ICP) algorithm is then used for point cloud registration and difference analysis to calculate the root mean square error (RMSE) between the new model and the baseline model. If the RMSE > 5 mm, the deformation is confirmed.

[0046] 3. Closed-loop expert assessment and handling Experts log into the system via the web interface, retrieve historical data and environmental data (such as rainfall curves) for comprehensive analysis, and input disposal measures (such as "strengthening drainage").

[0047] The system tracks the treatment effect and continuously monitors the settlement rate until it is below 0.05 mm / day for 24 consecutive hours. After the maintenance personnel confirm, the work order is closed, forming a closed-loop management record.

[0048] Example 4: System Self-Learning and Optimization Data-driven optimization: The system regularly (e.g., monthly) reviews historical alarm events and their handling results. False alarm / missed alarm analysis: False alarm data are marked as negative samples to optimize alarm thresholds (e.g., adjusting the settlement rate threshold to 0.15 mm / day).

[0049] LSTM model retraining: Retraining the model after incorporating new data to improve prediction accuracy.

[0050] Dynamic threshold adjustment: Automatically optimizes multi-level alarm thresholds based on long-term monitoring data. For example, in soft soil areas, the secondary alarm rate threshold is dynamically adjusted to 0.15 mm / day; the Pearson correlation coefficient r between settlement rate and rainfall is calculated, and a correlation warning is triggered when |r|>0.7.

[0051] Through the specific implementation methods described above, the intelligent substation settlement inspection system and method based on a three-dimensional model of the present invention ensures accurate binding between multi-source data and the model through a dual association and cross-validation mechanism, solving the data silo problem; it realizes the transformation from passive alarm to active prediction through LSTM neural network prediction, reducing the risk incidence rate; it makes the settlement status clear at a glance through dynamic rendering technology, improving decision-making efficiency; it reduces manual intervention through full-process automated management, improving response speed and handling effect; and it continuously improves system performance based on a self-learning mechanism of historical data, ensuring long-term reliability.

[0052] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A substation intelligent settlement inspection system based on a three-dimensional model, characterized in that, include: The sensing layer, used to collect settlement monitoring data, consists of a multi-source heterogeneous sensor network deployed in the substation; the multi-source heterogeneous sensor network includes a hydrostatic level, a GNSS displacement monitoring station, an unmanned aerial vehicle (UAV) aerial photography system, a fixed monitoring camera, and an inspection robot. The network layer is used to implement the encrypted transmission and aggregation of the monitoring data from the perception layer; The platform layer, as the core processing unit of the system, is used to fuse, analyze, visualize, and provide early warnings for monitoring data. The application layer provides users with a visual access interface; The platform layer further includes: ① The data fusion and mapping module is configured as follows: a. Associate fixed sensor data with corresponding model components through a predefined sensor ID and BIM model component coding mapping table; b. For GNSS and UAV data, dynamically bind them to the nearest vertex of the real-world 3D model using a spatial nearest neighbor query algorithm; c. Perform cross-validation on the two correlation results mentioned above, and trigger a calibration alarm when the validation results are inconsistent; ② Intelligent settlement analysis module, which integrates a prediction unit based on LSTM neural network, takes historical settlement time series data as input and outputs future settlement amount and probability of exceeding limit; ③ A 3D visualization engine is used to integrate lightweight BIM models with real-world models and to achieve dynamic rendering of settlement data through color mapping and vertex displacement technology. ④ The early warning and response management module is used to automatically generate work orders, dispatch drones for review, and track and handle closed loops.

2. The intelligent substation settlement inspection system based on a three-dimensional model according to claim 1, characterized in that, The early warning and response management module adopts a multi-level alarm mechanism, which is configured as follows: Level 1 alarm: Triggered when the cumulative settlement exceeds the design allowable value and the duration exceeds the first set threshold; Level 2 alarm: Triggered when the settlement rate exceeds 0.2 mm / day and exceeds the limit for three consecutive sampling cycles; Level 3 warning: triggered when the probability of settlement exceeding the limit within a future set time period output by the LSTM neural network prediction unit is greater than the probability threshold.

3. The intelligent substation settlement inspection system based on a three-dimensional model according to claim 1, characterized in that, The spatial nearest neighbor query algorithm specifically calculates the Euclidean distance between the sensor coordinates and the vertices of the triangulation network of the real scene model, and binds them to the vertex or virtual measurement point with the smallest distance.

4. The intelligent substation settlement inspection system based on a three-dimensional model according to claim 1, characterized in that, The 3D visualization engine employs instantiation rendering and LOD technology, and reduces the number of model faces to below 5 million through the Quadric Mesh Simplification algorithm.

5. The intelligent substation settlement inspection system based on a three-dimensional model according to claim 1, characterized in that, After triggering an alarm, the early warning and response management module automatically instructs the UAV to conduct aerial photography verification and uses the ICP algorithm to perform point cloud difference analysis to confirm the deformation.

6. The intelligent substation settlement inspection system based on a three-dimensional model according to claim 1, characterized in that, The data processing at the platform layer includes sliding window skipping, Kalman filtering smoothing, and time series interpolation to ensure data continuity.

7. A method for intelligent settlement inspection of substations based on a three-dimensional model, characterized in that, When applied to the system according to any one of claims 1 to 6, the method includes the following steps: S1: Settlement data is collected through a multi-source sensor network; S2: Dynamically bind data to the 3D model through a dual association and cross-validation mechanism; S3: Predicting settlement trends based on LSTM neural networks; S4: Perform dynamic visualization rendering in the merged 3D model; S5: Triggers multi-level alarms and initiates closed-loop handling procedures.

8. The intelligent settlement inspection method for substations based on a three-dimensional model according to claim 7, characterized in that, The dual association and cross-validation mechanism includes: Link fixed sensors to BIM components using a mapping table; Dynamic monitoring data is linked to real-world models using spatial coordinates; The two correlation results are compared and calibrated in real time.

9. The intelligent settlement inspection method for substations based on a three-dimensional model according to claim 7, characterized in that, The dynamic visualization rendering includes: Dynamically generate color mapping based on alarm thresholds; Deformation is simulated by displacing the vertices of the BIM model mesh using a vertex shader.

10. The intelligent settlement inspection method for substations based on a three-dimensional model according to claim 7, characterized in that, The closed-loop processing procedure includes: Automatically generate work orders and dispatch them to maintenance personnel; Deformation verification was performed using drone aerial photography data. Record the measures taken and verify their effectiveness until the alarm is cleared.

Images