Visual decision-making method and device for multi-source data of digital twin substation and medium

Abstract

The invention discloses a digital twin substation multi-source data visualization decision-making method, which comprises the following steps: constructing a three-dimensional holographic digital twin model of substation equipment, integrating BIM data and GIS geographic information, and carrying out lightweight processing; multi-source heterogeneous data, including equipment state data, environment sensing data and video monitoring data, of the transformer substation are collected in real time. Through the technologies of multi-source data fusion, dynamic digital twin modeling, AI aid decision making and the like, holographic data integration is realized, data islands are broken, and unified analysis and visualization of multi-dimensional data such as equipment states, environments, videos and the like are realized. And real-time dynamic mapping is carried out: high-precision digital twin bodies are constructed, the operation state of a physical substation is synchronized, and the fault positioning and prediction capability is improved. And intelligent decision support: in combination with machine learning and an expert knowledge base, fault root cause analysis, risk assessment and optimization operation and maintenance schemes are provided, and manual intervention requirements are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grids and digital substations, and specifically relates to a method, device and medium for visualizing decision-making of multi-source data in a digital twin substation. Background Art

[0002] Currently, existing technologies in the field of intelligent substations mainly include traditional SCADA monitoring systems, single data visualization platforms, and offline simulation systems. However, these technologies still have the following drawbacks:

[0003] Serious data silos: Existing systems often use independent subsystems (such as equipment monitoring, environmental sensing, and video surveillance), with dispersed data storage and a lack of unified integrated analysis, leading to delayed decision-making. For example, transformer oil chromatography data and temperature monitoring data are not linked, making it difficult to accurately predict faults.

[0004] Insufficient visualization capabilities: Existing 3D visualization is mostly based on static modeling, which cannot map the physical substation status in real time. It also lacks the overlay analysis of multi-source data (such as infrared thermal imaging and partial discharge monitoring), affecting operation and maintenance efficiency.

[0005] Weak decision support: Traditional systems rely on manual experience and judgment, and AI algorithms are only used for single indicator warnings (such as overtemperature alarms). They do not combine historical equipment data and environmental factors for comprehensive risk assessment, resulting in a high false alarm rate.

[0006] Simulation is disconnected from operation and maintenance: Digital twin technology is mostly used in the design phase and is not deeply integrated with real-time operation and maintenance data, resulting in delayed updates of simulation models and the inability to dynamically optimize operation strategies. Summary of the Invention

[0007] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a digital twin substation multi-source data visualization decision method, device and medium to solve the problems in the above-mentioned background technology.

[0008] The object of the present invention is achieved as follows: a digital twin substation multi-source data visualization decision method, which includes the following steps:

[0009] Build a 3D holographic digital twin model of substation equipment, integrate BIM data with GIS geographic information, and perform lightweight processing;

[0010] Real-time collection of multi-source heterogeneous data from substations, including equipment status data, environmental perception data, and video surveillance data;

[0011] Establish the topological relationship between devices, sensors and events through the graph database to generate a dynamic association graph;

[0012] Use AI models to integrate and analyze multi-source data, predict equipment degradation trends, and generate optimization decisions;

[0013] Render multi-source data in layers on a 3D visualization platform to achieve multi-dimensional interactive presentation of device status and decision feedback.

[0014] The construction of the three-dimensional holographic digital twin model includes: generating a detailed equipment-level model based on a BIM modeling tool and exporting topological relationships through the IFC format; superimposing GIS terrain data and using spatial mosaic technology to match the BIM model with the geographic coordinate system; and lightweight processing of the model, including geometric simplification, texture compression, and format conversion to glTF or FBX format.

[0015] The multi-source heterogeneous data is collected using edge computing nodes, including support for IEC61850, Modbus and MQTT protocols; data standardization processing includes timestamp alignment, noise filtering and real-time synchronization based on the OPC UA protocol, and is stored in a time series database.

[0016] Define node types as devices, sensors, electrical nodes, and events; build physical connections, electrical paths, and event trigger relationships; parse the adjacency matrix based on the single-line diagram to automatically generate physical connection relationships between devices.

[0017] The AI ​​model includes:

[0018] The LSTM neural network predicts the remaining service life of the equipment; the knowledge graph matches the historical failure case library to quantify the failure probability and risk weight; the multi-objective optimization model generates a Pareto optimal solution set, combining equipment life, operating efficiency and maintenance cost to make a trade-off.

[0019] The implementation of the layered rendering includes:

[0020] Infrared thermal imaging data is mapped to the device surface, and a dynamic thermal distribution map is generated through GPU shaders. A particle system is used to visualize partial discharge signals, with the intensity driven by the discharge quantity parameter. Touch screen interaction is supported, and device-related data, including vibration spectra and oil chromatogram historical trends, can be retrieved through X-ray detection.

[0021] The generation of the optimization decision is specifically as follows:

[0022] When the predicted equipment degradation threshold exceeds the limit, maintenance recommendations are automatically pushed and spare parts allocation orders are generated; the transformer operation strategy is adjusted according to real-time load data to balance equipment life and energy efficiency; historical fault scenarios are replayed through the digital twin model to optimize the protection setting.

[0023] A digital twin substation multi-source data visualization decision-making device, comprising:

[0024] Twin modeling module, used to build lightweight 3D models and achieve dynamic data synchronization;

[0025] Data fusion module, used for real-time collection, standardization and graph database storage of multi-source heterogeneous data;

[0026] AI decision-making module for equipment status prediction, risk assessment, and strategy optimization;

[0027] Visual interaction module for multi-layer rendering, touch screen operation and decision feedback.

[0028] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method when executing the computer program.

[0029] A computer-readable storage medium includes a plurality of program codes, and the program codes are used to be loaded and run by a processor to execute the above method.

[0030] The beneficial effects of the present invention: Through multi-source data fusion, dynamic digital twin modeling, AI-assisted decision-making and other technologies, the following are achieved: Holographic data integration: breaking down data silos to achieve unified analysis and visualization of multi-dimensional data such as equipment status, environment, and video. Improve visualization capabilities, real-time dynamic mapping: build high-precision digital twins, synchronize the operating status of physical substations, and improve fault location and prediction capabilities. Intelligent decision support: combining machine learning with expert knowledge base to provide root cause analysis, risk assessment, and optimized operation and maintenance solutions, reducing the need for manual intervention. Closed-loop optimization: through the interaction of twin models with real-time data, dynamically adjust the operating strategy, extend equipment life and improve grid reliability, and realize intelligent management of the substation throughout its life cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] none DETAILED DESCRIPTION

[0032] The present invention is further described below in detail, which does not limit the present invention but is only for the purpose of more clearly illustrating and explaining the present invention.

[0033] This embodiment discloses a method for visualizing and making decisions based on multi-source data of a digital twin substation, including the following steps:

[0034] Build a 3D holographic digital twin model of substation equipment, integrate BIM data with GIS geographic information, and perform lightweight processing;

[0035] Real-time collection of multi-source heterogeneous data from substations, including equipment status data, environmental perception data, and video surveillance data;

[0036] Establish the topological relationship between devices, sensors and events through the graph database to generate a dynamic association graph;

[0037] Use AI models to integrate and analyze multi-source data, predict equipment degradation trends, and generate optimization decisions;

[0038] Render multi-source data in layers on a 3D visualization platform to achieve multi-dimensional interactive presentation of device status and decision feedback.

[0039] (1) Multi-source data fusion: edge computing nodes are used to pre-process heterogeneous data (IoT sensors, SCADA, video streams), achieve real-time access through a unified data bus (such as Apache Kafka), and use time series databases (such as InfluxDB) to store structured data.

[0040] Part 1: IoT sensor data preprocessing:

[0041] Objectives: denoising, anomaly detection, downsampling and normalization.

[0042] 1) Data cleaning

[0043] Sliding average filter:

[0044] Use a moving average with a window size of N (such as 10) to smooth high-frequency noise. This is suitable for slowly varying signals such as temperature and voltage.

[0045]

[0046] N represents the window size, that is, the number of consecutive data points used in each calculation of the average value, y t represents the filtered output value (smoothed result) at time point t, x t-i Represents the value of the original input signal at time point ti (i ranges from 0 to N-1).

[0047] Kalman filter:

[0048] Model dynamic systems (such as vibration sensors) and iteratively optimize estimates through prediction-update:

[0049]

[0050] in represents the optimal state estimate at time t, represents the prior state prediction at time t, Kt is the Kalman gain, which weighs the model prediction and the observation, Zt represents the actual observation at time t, and H represents the matrix that maps the system state to the observation space.

[0051] 2) Anomaly Detection

[0052] Z-Score method

[0053] Calculate the Z value of the data within the window and mark the points that exceed the threshold (such as ±3) as abnormal:

[0054]

[0055] Where Z represents the standardized score of the current data point, indicating the number of standard deviations from the mean. χ is the original value of the current data point (to be tested for anomalies). μ represents the mean of the data within the window (e.g., the average of the most recent 10 data points). σ represents the standard deviation of the data within the window, which measures the degree of data dispersion. Outliers are replaced using linear interpolation or the mean of the previous and next values.

[0056] Dynamic threshold method:

[0057] Set adaptive thresholds based on historical data in different time periods (such as peak and valley power consumption periods) to avoid misjudgment of fixed thresholds.

[0058] 3) Data compression

[0059] Segmented Aggregate Approximation (PAA): The data is segmented into k intervals, each segment is represented by the mean, and the compression ratio is adjustable (such as 10:1).

[0060] Part 2: SCADA system data preprocessing

[0061] Objectives: protocol parsing, time alignment, event extraction.

[0062] 1) Protocol conversion:

[0063] Convert Modbus / OPC UA to JSON, parse SCADA original messages, and map fields to a unified JSON format:

[0064]

[0065] 2) Time synchronization

[0066] NTP / PTP clock synchronization: Use Network Time Protocol (NTP) or Precision Time Protocol (PTP) to align edge node and SCADA clocks with an error of <1ms.

[0067] 3) Event Detection

[0068] State transition rule engine: Define rules (such as circuit breaker tripping) to trigger events:

[0069] if(current>threshold)and(status=="closed"):

[0070] trigger_alert("overcurrent_trip")

[0071] Part 3: Video Stream Data Preprocessing

[0072] Objectives: keyframe extraction, feature compression, and metadata generation.

[0073] 1) Frame sampling and compression

[0074] Adaptive frame rate sampling:

[0075] Dynamically adjust the frame rate (e.g. 5fps in normal state, 25fps when motion is detected) to save bandwidth.

[0076] H.265 encoding:

[0077] Uses hardware-accelerated H.265 compression with a compression ratio of 50:1.

[0078] 2) Feature extraction

[0079] YOLOv5 object detection: Lightweight model to detect device status (such as dashboard readings, smoke):

[0080] model=torch.hub.load('ultralytics / yolov5','yolov5s')

[0081] results=model(frame)

[0082] Background subtraction (ViBe algorithm): Detects dynamically changing areas and transmits only pixels in the difference area to reduce the amount of data.

[0083] Metadata Generation:

[0084] Structured log: Output detection results to JSON:

[0085]

[0086] Part 4: Data Fusion and Transmission

[0087] Kafka topic partitioning strategy: Topics are divided by data type (such as sensors, scada, and video). Sensor data is partitioned by device ID hash to ensure time sequence order.

[0088] Edge cache queue: Enables local SQLite cache when the network is interrupted and retransmits after recovery.

[0089] Part 5: Time Series Data Storage Optimization

[0090] InfluxDB Schema Design:

[0091] Tags: device_id, region

[0092] Fields: value, status

[0093] Timestamp: nanosecond precision, supports fast range queries.

[0094] Through the above preprocessing, edge nodes achieve data lightweighting and standardization, providing real-time data streams with high signal-to-noise ratio for subsequent visualization and decision-making, while significantly reducing the load on central servers.

[0095] Working principle: After the data is standardized (such as IEC 61850 protocol conversion), feature extraction and association analysis (such as device topology association based on a graph database) are used to construct a data association map for the entire site.

[0096] Step 1: Feature Extraction

[0097] (1) Time series data feature extraction

[0098] Statistical features: Extract sliding window features from sensor / SCADA time series data:

[0099] The Python code is as follows:

[0100] #Example: 5-minute window features

[0101]

[0102] Event features: Detect abnormal events through LSTM-Autoencoder and extract potential feature vectors: The Python code is as follows:

[0103]

[0104] (2) Device attribute characteristics

[0105] Structured metadata: parses device attributes from IEC 61850 SCL (Substation Configuration Description Language). Its XML format is as follows:

[0106]

[0107]

[0108] Extract key fields: device ID, voltage level, bay, and control logic node.

[0109] Status tag generation: Generate device health tags based on SCADA status: health_score = 0.7*normalized(temperature) + 0.3*normalized(vibration) Step 2: Association modeling

[0110] (1)Graph model design

[0111] Node types and attributes:

[0112] Node Type Attribute Example Device id, name, type, voltage_level Sensor id, metric_type, sampling_rate ElectricalNode id, phase(A / B / C / N), coordinates Event id, type, severity, timestamp

[0113] Relationship types and attributes:

[0114] Relationship Start Point End Point Attribute PHYSICALLY_CONNECTED Device Device cable_type, length BELONGS_TO Sensor Device installation_date ELECTRICAL_PATH Device ElectricalNode impedance TRIGGERS Event Device confidence CORRELATED_WITH Event Event correlation_coefficient

[0115] (2) Topological association rules

[0116] Physical connection inference: Analyze the connection matrix based on the single line diagram (SLD):

[0117] The Python code is as follows:

[0118]

[0119] Automatically generate PHYSICALLY_CONNECTED relationships.

[0120] Electrical coupling analysis: implicit associations are established through the power flow calculation matrix:

[0121] Y_bus=[[Y11,Y12,...],#node admittance matrix

[0122] [Y21,Y22,...], ...]

[0124] #Add ELECTRICAL_COUPLING relationship when Yij's modulus value > threshold

[0125] Step 3: Dynamic Correlation Analysis

[0126] (1) Real-time event propagation

[0127] Cypher-based causal reasoning:

[0128] / / Query the upstream device that caused the overtemperature alarm

[0129] MATCH(e:Event{type:"OverTemperature"})<-[:TRIGGERS]-(d:Device)

[0130] WITH d MATCH path=(d)-[:ELECTRICAL_PATH*..5]->(upstream:Device)

[0131] WHERE upstream.health_score<0.8

[0132] RETURN upstream.id,relationships(path)

[0133] Dynamic weight update: adjust relationship weights based on real-time data:

[0134] The Python code is as follows:

[0135]

[0136] (2) Application of graph algorithms

[0137] Community detection (Louvain algorithm), identifying clusters of tightly electrically coupled devices:

[0138] Shortest fault path analysis, using Dijkstra algorithm to locate the minimum impedance fault path:

[0139]

[0140]

[0141] Step 4: Graph Storage Optimization

[0142] Indexing strategy:

[0143] CREATE INDEX ON:Device(id); / / Device ID index

[0144] CREATE INDEX ON:Event(timestamp); / / Time range index

[0145] Time-sliced ​​storage:

[0146] Store graph state snapshots as time-series subgraphs on an hourly basis, supporting historical backtracking:

[0147] / graph_202310011200 / / 12:00 subgraph version

[0148] / graph_202310011300 / / 13:00 subgraph version

[0149] (2) Dynamic digital twin modeling

[0150] Solution: Build a high-precision 3D model based on BIM+GIS, and implement web-based visualization using the lightweight rendering engine Three.js. Use real-time data streams to drive model status updates (such as color gradients triggered by transformer temperature changes).

[0151] Working principle: Physical devices and virtual models communicate bidirectionally through OPC UA or MQTT protocols. The twin is dynamically adjusted based on real-time data and supports historical data playback and simulation deduction.

[0152] 1) BIM+GIS data fusion

[0153] Build a detailed model of the substation equipment level using BIM modeling tools (such as Revit), and export topological relationships and attribute data in IFC format;

[0154] Overlay GIS geographic coordinate system data (such as terrain, roads, and pipelines) and use spatial mosaic technology to achieve accurate matching between the BIM model and GIS terrain;

[0155] Use 3D tile specifications (such as S3M) to divide the model into blocks, support multiple levels of detail (LOD) and instanced storage, and reduce memory usage.

[0156] 2) Lightweight processing

[0157] Geometry simplification: Reduce model complexity through mesh merging, redundant vertex culling, and triangle face optimization (such as Unity's Mesh Combine function);

[0158] Texture compression: Convert high-resolution textures to WebP or KTX2 format and use mipmap chains to adapt to different zoom levels;

[0159] Data format conversion: Convert BIM models to glTF or FBX format, retaining materials, animations, and hierarchical structures, and adapting to the Three.js / Unity3D engine.

[0160] 3) Implementing a lightweight rendering engine based on Three.js

[0161] Scene initialization:

[0162] Create a scene (THREE.Scene), a camera (THREE.PerspectiveCamera), and a WebGL renderer; load a glTF model file, parse it using GLTFLoader, and add it to the scene; set material lighting parameters (such as THREE.StandardMaterial).

[0163] Interaction and optimization:

[0164] Viewpoint control: realize translation, rotation and scaling of the model through OrbitControls;

[0165] Dynamic update: Use shaders (ShaderMaterial) to achieve color gradient effects for temperature data, and pass real-time data through uniform variables;

[0166] Performance optimization: Use instanced rendering (InstancedMesh) to handle repeated equipment (such as insulators) to reduce DrawCall.

[0167] 4) Key technologies for performance optimization

[0168] LOD dynamic loading:

[0169] Switch the model detail level based on the camera distance, displaying simplified meshes at long distances and loading high-precision models at close ranges;

[0170] In Three.js, multi-level models are managed through LOD objects, and Unity uses LOD Group components.

[0171] Data chunking and on-demand loading:

[0172] The substation is divided into modules such as the electrical area and the control room, and the visible area is dynamically loaded using the octree spatial index;

[0173] Combined with Service Worker to preload adjacent blocks to reduce interaction lag.

[0174] GPU-accelerated computing:

[0175] Use GPGPU technology (such as GPUComputationRenderer) in Three.js to process device state data in parallel;

[0176] Unity uses Compute Shader to achieve real-time mapping of large-scale data (such as temperature field simulation).

[0177] (3) AI-assisted decision-making engine

[0178] Solution: Integrate deep learning (LSTM to predict equipment degradation) and knowledge graph (fault case library) to build a multi-objective optimization model (such as balancing equipment life and operating efficiency).

[0179] Working principle: After real-time data is input into the decision engine, expert experience is integrated through federated learning to output failure probability, maintenance priority and optimization strategy (such as load adjustment suggestions).

[0180] 1) Multi-objective optimization model framework:

[0181] The model is based on real-time data (such as equipment status and environmental parameters) of the digital twin substation and a historical fault case library, combined with LSTM-predicted equipment degradation trends, to construct a multi-objective optimization problem. Its core structure includes:

[0182] Input: Remaining useful life (RUL) predicted by LSTM, fault association rules in the knowledge graph, and real-time operating parameters (temperature, current, vibration, etc.).

[0183] Output: Optimized equipment operation strategy (such as load distribution, maintenance plan), and trade-off between equipment life and efficiency.

[0184] 2) Objective function design:

[0185] The objective function needs to reflect the core optimization requirements of substation operation. Common objectives include:

[0186] Maximize equipment life:

[0187] Based on the remaining useful life (RUL) predicted by LSTM, optimize the equipment load distribution to reduce the degradation rate. For example:

[0188]

[0189] Among them, α i is the equipment weight coefficient, reflecting its importance in the system. RUL is the remaining useful life of the equipment, and n is the target number of equipment.

[0190] Maximize operational efficiency:

[0191] Optimize the overall energy efficiency of the substation, such as reducing line losses or increasing load factors:

[0192]

[0193] Where η represents efficiency, P output Indicates the effective power actually output by the system (such as mechanical work, electrical energy, thermal energy, etc.), P input Indicates the total input power (input electrical energy) obtained by the system from the outside.

[0194] Minimize maintenance costs:

[0195] Combined with failure cases in the knowledge graph, maintenance needs can be predicted and costs can be optimized:

[0196]

[0197] Among them, C j is the maintenance operation cost, x j is the decision variable (whether to perform maintenance).

[0198] Minimize risk index:

[0199] By associating failure modes through knowledge graphs, potential risks can be quantified. For example, based on historical failure probabilities:

[0200]

[0201] R k is the severity weight of the failure consequence.

[0202] 3) Constraints

[0203] Constraints must cover physical limitations of the device, safety regulations, and empirical rules from the knowledge graph:

[0204] Equipment operating parameter constraints:

[0205] T i ≤T max , I j ≤I rated

[0206] If the transformer temperature T i Must not exceed the threshold value, current I j Do not exceed rated values.

[0207] Maintenance cycle constraints:

[0208] According to the fault case library in the knowledge graph, set the minimum maintenance interval:

[0209] t last_maintenance +△t≤t current

[0210] Resource limitations:

[0211] Such as budget, spare parts inventory, manpower, etc.:

[0212]

[0213] B j This is the resource consumption for a single maintenance.

[0214] Fault avoidance constraints:

[0215] Based on the causal rules in the knowledge graph, high-risk operation combinations are prohibited. For example, if historical data shows that "overload + high temperature easily causes insulation failure", then:

[0216] \text{if}P_{\text{load}}\geq P_{\text{threshold}}\text{and}T\geqT_{\text{threshold}},\text{then}x_{\text{overload}}=0$$[11,14](@ref)

[0217] 4). Model solution method

[0218] Multi-objective optimization algorithm:

[0219] Genetic algorithm (NSGA-II), particle swarm optimization (MOPSO) and other algorithms are used to generate Pareto optimal solution sets, providing multiple trade-off options for decision makers to choose from.

[0220] Knowledge graph assists decision making:

[0221] Match the Pareto solution set with the failure cases in the knowledge graph and give priority to solutions with low historical risks.

[0222] 5) Practical Application Examples

[0223] Take transformer load distribution as an example:

[0224] Input: LSTM predicts that the remaining life of transformer A is 2 years, and the knowledge graph shows that its historical overload failure probability is 15%.

[0225] Goal: Trade off between lifespan (Goal 1) and efficiency (Goal 2).

[0226] Output: Reduce the load factor of transformer A to 80%, extend the remaining life to 2.5 years, and control the efficiency loss within 5%.

[0227] (4) Interactive visualization platform

[0228] Solution: Use layered rendering technology to overlay infrared thermal imaging, partial discharge signals and other data layers on the 3D model to support touch-screen interactive operations.

[0229] Working principle: Multi-source data is anchored to the device model through spatial coordinate matching. Users can retrieve related data (such as vibration spectrum and oil chromatography historical trends) by clicking on the device.

[0230] The layered rendering process dynamically overlays multi-source data (such as infrared thermal imaging and partial discharge signals) with the 3D model to achieve multi-dimensional visualization and interactive operation of the device status. The specific process is as follows:

[0231] 1. Data preprocessing and hierarchical construction

[0232] Data source classification: Different monitoring data (infrared thermal images, partial discharge signals, vibration spectra, etc.) are divided into independent data layers according to their types, and each layer corresponds to a specific physical quantity or equipment status parameter.

[0233] Data cleaning and alignment: De-noise, filter, and standardize the raw data to ensure that the timestamps of each layer of data are consistent with the spatial coordinate system (such as the WGS84 coordinate system or the device local coordinate system).

[0234] Layer attribute definition: Set rendering attributes for each layer (such as transparency, color mapping rules, and dynamic update frequency). For example, the infrared thermal imaging layer uses a red-yellow-blue gradient to represent the temperature gradient, and the partial discharge signal layer uses flashing light spots to mark the discharge intensity.

[0235] 2. Spatial coordinate matching and data anchoring

[0236] 3D model benchmark construction: Generate a high-precision 3D model of the device through lidar scanning or oblique photography technology, and establish a unified device space coordinate system.

[0237] Multi-source data spatial registration: Utilizes feature point matching algorithms (such as the ICP algorithm) to anchor each data layer to the corresponding location on the model. For example, infrared thermal imaging data is matched to the model's geometric vertices using device surface temperature sampling points, and partial discharge signals are mapped to the insulator model using the three-dimensional coordinates of the discharge location.

[0238] Dynamic data synchronization: Real-time collected data is synchronized with the model animation frame through timestamps to ensure that the data layer is dynamically consistent with the device operating status.

[0239] 3. Layered rendering and overlay fusion

[0240] Rendering engine selection: Use an engine that supports multi-channel rendering, such as Unreal Engine 5 (UE5) or Unity, to achieve high frame rate and high-fidelity rendering.

[0241] Independent rendering pass: Each layer of data is assigned an independent rendering pass, for example:

[0242] Basic geometry layer: basic mesh and material rendering of the device's 3D model.

[0243] Thermal imaging layer: Generates thermal distribution maps in real time based on GPU shaders and overlays them on the device surface.

[0244] Partial discharge signal layer: simulate the discharge position and intensity through the particle system, and dynamically adjust the particle size and color.

[0245] Transparency blending and occlusion processing: Use Alpha blending technology to control the overlay effect of each layer, and enable depth testing to avoid visual occlusion (for example, the partial discharge signal layer is preferentially displayed above the device casing).

[0246] 4. Interaction and data association

[0247] Touch screen event response: Raycasting technology is used to identify the device component corresponding to the user's click location and trigger the retrieval of related data.

[0248] Dynamic Data Panel: After clicking on a device, relevant information such as vibration spectrum historical trend (time series data visualization) and oil chromatography analysis report (chart and text) is called up and presented in the form of a floating window or sidebar.

[0249] Layer visibility control: supports user-defined display / hiding of specific data layers, such as superimposing only the infrared layer or displaying both the vibration and partial discharge layers simultaneously.

[0250] 5. Real-time updates and performance optimization

[0251] Streaming data transmission: Use edge computing technology to compress and transmit sensor data in real time to reduce rendering latency.

[0252] Dynamic LOD (Level of Detail) adjustment: Dynamically adjust the data layer resolution based on the user's viewing distance (such as reducing the sampling density of the thermal imaging layer at long distances).

[0253] GPU parallel computing: Use CUDA or OpenCL to accelerate data layer rendering, ensuring a smooth rendering of more than 60FPS when multiple layers are superimposed.

[0254] (1) Holographic data fusion: The first graph database-based substation data correlation analysis to achieve deep coupling of equipment, environment, and operation and maintenance data.

[0255] (2) Dynamic twin synchronization: A two-way update mechanism of "data-driven + model calibration" is proposed to solve the industry problem of lagging traditional digital twin models.

[0256] (3) Intelligent decision-making closed loop: Feedback AI prediction results to the twin model to form a closed-loop management of "monitoring-early warning-optimization" (such as automatically generating inspection routes).

[0257] (4) Interactive visualization: Develop multi-layer dynamic overlay technology to support operation and maintenance personnel to query the full life cycle data of the equipment with one click.

[0258] Deployment phase: Deploy a data acquisition gateway at the edge of the substation and build a twin platform.

[0259] The platform functional modules include:

[0260] Panoramic roaming: roam around the substation from a first-person perspective;

[0261] Equipment analysis: Displays the three-dimensional inspection points of a single power equipment, as well as historical data, current status, and trend data analysis of the inspection points;

[0262] Alarm Center: Displays historical alarm data, further displays the 3D alarm points of individual devices, as well as real-time video and historical patrol data of the points;

[0263] Remote inspection: display historical inspection tasks, robot inspection tasks, inspection record viewing, etc.

[0264] Operation and maintenance stage: Operation and maintenance personnel view the 3D visualization interface through the web terminal and receive warning information and decision-making suggestions pushed by the system (such as "the casing temperature is abnormal, it is recommended to shorten the inspection cycle").

[0265] Optimization stage: Based on historical data and simulation results, the patrol frequency, algorithm sensitivity, patrol scenarios, etc. are adjusted and optimized.

[0266] Through multi-source data fusion, dynamic digital twin modeling, AI-assisted decision-making and other technical means, significant breakthroughs have been achieved in the field of intelligent substation operation and maintenance. The specific beneficial effects are as follows:

[0267] 1. Improve data integration and analysis capabilities

[0268] Technical features:

[0269] Edge computing nodes are used to collect and standardize multi-source heterogeneous data such as SCADA, online monitoring, and video surveillance in real time (based on the IEC 61850 protocol).

[0270] By building a correlation graph of equipment-environment-operation and maintenance data through a graph database, deep data coupling can be achieved.

[0271] Beneficial effects:

[0272] Breaking down data silos: Traditional systems require manual switching of multiple independent platforms (such as temperature monitoring and oil chromatography analysis). This invention achieves "one-screen integration" of multi-source data through a unified data bus, improving operation and maintenance efficiency by more than 60%.

[0273] Enhanced analysis accuracy: By combining equipment topology relationships (such as the linkage between the transformer and the cooling system), the root cause of anomalies can be accurately located (such as oil temperature increase caused by a cooling fan failure), reducing false alarm rates by 40%.

[0274] 2. Achieve dynamic synchronization and high-fidelity visualization of digital twins

[0275] Technical features:

[0276] Build lightweight 3D models based on BIM+GIS, and realize real-time data-driven connection between physical devices and virtual models through OPC UA / MQTT protocols.

[0277] Using layered rendering technology, data layers such as infrared thermal imaging and partial discharge signals are superimposed on the three-dimensional scene.

[0278] Beneficial effects:

[0279] Dynamic mapping with zero delay: The traditional digital twin model update cycle takes several hours. This invention is driven by real-time data streams, and the model response time is shortened to seconds (such as the circuit breaker trip status synchronization in seconds).

[0280] Panoramic visualization: Operation and maintenance personnel can intuitively view the multi-dimensional status of equipment (for example, click on the transformer model to directly retrieve temperature, vibration, and oil chromatography data), reducing the time to obtain key information by 70%.

[0281] 3. Intelligent decision support and closed-loop optimization

[0282] Technical features:

[0283] Combined with the knowledge graph (fault case library), it provides root cause analysis and generates optimization strategies (such as load adjustment and inspection route planning).

[0284] Beneficial effects:

[0285] Accurately predict faults: Traditional threshold alarms cannot identify potential risks (such as the initial signals of partial discharge in bushings). This invention uses AI models to predict faults 24-48 hours in advance, reducing the unplanned downtime rate of equipment by 35%.

[0286] Closed-loop operation and maintenance optimization: The system automatically pushes maintenance recommendations (such as "the insulation of the lightning arrester has decreased, it is recommended to replace it within two weeks") and feeds back the results to the twin model calibration parameters to form a continuous optimization closed loop.

[0287] 4. Improve interaction efficiency and user experience

[0288] Technical features:

[0289] It supports interactive operations on mobile terminals and enables “one-click penetration” query of multi-source data through spatial coordinate matching.

[0290] Provide historical data playback and simulation deduction functions.

[0291] One-click penetration query means that when a user clicks on a device model, the three-dimensional coordinates (x, y, z) of the contact point are calculated through radiographic detection. Then, based on a spatial hash table, related data layers (such as partial discharge signals and vibration spectra) within a 0.5-meter radius of that coordinate are quickly retrieved. The top five relevant datasets are then prioritized and returned. For historical data, oil chromatogram trend curves for the past 30 days are extracted from the time series database (InfluxDB), and a sliding window algorithm is used to detect abnormal fluctuations.

[0292] Lowering the operating threshold: Traditional systems require professional training to operate. This invention enables novice operation and maintenance personnel to quickly get started through natural interaction (such as gesture rotation model and voice query data).

[0293] Assisted accident review: By replaying historical data, the entire fault process can be restored (such as the action sequence of busbar protection caused by lightning strikes), shortening the accident analysis time by 50%.

[0294] Comprehensive benefit comparison (the present invention vs. the existing technology)

[0295]

[0296]

[0297] The method of the present invention is applicable to intelligent operation and maintenance of substations, including the following scenarios:

[0298] Real-time monitoring of equipment status: such as temperature, vibration, partial discharge and other parameter monitoring of key equipment such as transformers, circuit breakers and lightning arresters.

[0299] Fault prediction and diagnosis: AI analysis can be used to detect potential faults (such as insulation degradation and mechanical wear) in advance.

[0300] Operation and maintenance decision support: Automatically generate inspection plans, maintenance recommendations, and operation optimization strategies.

[0301] Emergency response and accident review: quickly locate the fault point, restore the accident process, and assist in formulating emergency repair plans.

[0302] Specific usage steps

[0303] Step 1: System deployment and data access

[0304] Edge-side deployment: Install intelligent data collection terminals (such as IoT sensors and video cameras) at substations to transmit data in real time to the cloud or local servers via 5G / fiber optic networks.

[0305] Data standardization: The system automatically converts the protocol (such as IEC61850) and cleans multi-source data such as online monitoring and environmental sensors, and stores them in a time series database.

[0306] Step 2: Digital twin modeling and dynamic synchronization

[0307] 3D modeling: Build a high-precision 3D model of the substation based on BIM+GIS and import it into a lightweight rendering engine (such as WebGL).

[0308] Real-time drive: Map physical device data to virtual models through OPC UA / MQTT protocols to achieve dynamic updates (such as a transformer temperature increase triggering a model color change).

[0309] Step 3: Multi-source data visualization and interaction

[0310] Data overlay: Infrared thermal imaging, partial discharge signals, vibration spectrum and other data are displayed in layers in the 3D model (for example, red areas indicate temperature anomalies).

[0311] Interactive operation: Operation and maintenance personnel can click on the device model through a PC or mobile terminal to directly view historical trends, related parameters and AI analysis reports.

[0312] Step 4: AI analysis and decision support

[0313] Fault warning: The system uses the LSTM model to predict equipment degradation trends (e.g., "the insulation performance of the bushing may drop to the threshold in the next 7 days").

[0314] Decision generation: Combined with the knowledge graph, it recommends optimization measures (such as "adjusting load distribution" or "prioritizing casing replacement") and pushes them to the mobile terminal of the operation and maintenance personnel.

[0315] Step 5: Closed-loop optimization and feedback

[0316] Strategy execution: Operations and maintenance personnel perform maintenance according to system recommendations, and the resulting data (such as temperature changes after maintenance) is fed back to the digital twin model.

[0317] Model calibration: The system automatically optimizes AI algorithm parameters to improve the accuracy of subsequent predictions.

[0318] Effects and benefits:

[0319] (1) Operation and maintenance efficiency is significantly improved

[0320] Traditional method: Manual switching of multiple systems is required, and it takes an average of 30 minutes to process a single alarm.

[0321] This invention: Through "one-screen fusion" visualization, abnormality positioning and analysis can be completed within 5 minutes, improving efficiency by 80%.

[0322] (2) Enhanced fault prediction capabilities

[0323] Traditional method: relying on regular inspections, with a missed inspection rate of up to 25%.

[0324] This invention: AI predicts failures 24-48 hours in advance, reducing unplanned outages by 35%.

[0325] (3) Improved scientific nature of decision-making

[0326] Traditional method: relying on experience and judgment, with an error rate of about 20%.

[0327] This invention: Based on multi-dimensional data and knowledge graphs, the decision-making accuracy is increased to 95%.

[0328] (4) Dual optimization of cost and safety

[0329] Economical: Preventive maintenance reduces emergency repair costs and reduces annual operation and maintenance costs by 25%.

[0330] Safety: Real-time monitoring of high-risk equipment (such as oil-filled equipment) reduces the risk of accidents such as fire by 50%.

[0331] Typical application cases

[0332] Case 1: Transformer overheating warning

[0333] Phenomenon: The system detected an abnormally high transformer oil temperature. AI analysis combined with historical data determined that the cooling fan was faulty.

[0334] Action: Automatically push the "replace fan" suggestion and generate a spare parts transfer order.

[0335] Result: Transformer overload damage was avoided, saving over 500,000 yuan in maintenance costs.

[0336] Case 2: Lightning Strike Accident Review

[0337] Phenomenon: Lightning strike causes busbar protection to malfunction.

[0338] Action: Use digital twins to replay voltage fluctuations and protection action timing at the moment of lightning strike.

[0339] Result: The positioning protection constant value setting defect was found, and similar accidents were eliminated after optimization.

[0340] Edge computing-based data collection and standardization technology (supporting multi-protocol compatibility such as IEC 61850 and Modbus). Graph database-driven topological correlation of equipment-environment-operation and maintenance data (e.g., dynamic coupling of transformer temperature and cooling system status). This breaks through the limitations of traditional system data silos and achieves "one-graph integration" for all site data.

[0341] Lightweight BIM+GIS 3D modeling technology (model face count reduced to 30% of traditional solutions). A real-time data-driven mechanism based on OPCUA / MQTT (physical devices and virtual models synchronized within seconds). This solves the traditional digital twin model update lag issue and achieves "zero-latency mapping" for the first time.

[0342] A fault prediction model combining LSTM and knowledge graphs (with >95% accuracy). Federated learning-driven O&M strategy optimization (such as a load distribution and equipment lifespan balance algorithm). An upgrade from a "single-point alarm" to a "prediction-decision-feedback" closed loop.

[0343] Three-dimensional spatial coordinate matching algorithm (pixel-level alignment of infrared thermal imaging and device models). Natural interaction technology for mobile terminals (gesture recognition, voice query). For the first time, immersive "what you see is what you get" operation has been achieved in substation operation and maintenance.

[0344] The multi-source data fusion system for substations based on digital twins includes an edge acquisition layer, a data bus layer, a twin modeling layer, and an AI decision-making layer (to protect the overall architecture). Edge computing nodes are used for multi-protocol data acquisition; a time series database stores standardized data; a three-dimensional twin engine drives model updates through real-time data streams; and an AI decision-making module outputs fault prediction and optimization strategies.

[0345] The LSTM-knowledge graph hybrid model for equipment degradation prediction and the multi-source data spatiotemporal alignment algorithm include acquiring multi-dimensional time series data of equipment; calculating degradation trends through the LSTM model; matching historical failure cases with the knowledge graph; and outputting root cause analysis and maintenance priorities.

[0346] Dynamic multi-layer overlay technology within 3D scenes and touchscreen-based O&M interaction methods include: anchoring infrared thermal imaging data within 3D models; accessing associated parameters by clicking on a device through spatial coordinate mapping; and support for gesture rotation and voice queries. A graph-based substation data correlation analysis method includes: constructing a device topology diagram; calculating the impact of environmental factors on device status using edge weights; and generating a station-wide health score.

[0347] A digital twin substation multi-source data visualization decision-making device, comprising:

[0348] Twin modeling module, used to build lightweight 3D models and achieve dynamic data synchronization;

[0349] Data fusion module, used for real-time collection, standardization and graph database storage of multi-source heterogeneous data;

[0350] AI decision-making module for equipment status prediction, risk assessment, and strategy optimization;

[0351] Visual interaction module for multi-layer rendering, touch screen operation and decision feedback.

[0352] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method when executing the computer program.

[0353] A computer-readable storage medium includes a plurality of program codes, and the program codes are used to be loaded and run by a processor to execute the above method.

[0354] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes based on the technical solutions and concepts of the present invention within the technical scope disclosed by the present invention, and they should be covered by the scope of protection of the present invention.

Claims

1. A digital twin substation multi-source data visualization decision-making method, characterized by: The following steps are involved: Build a 3D holographic digital twin model of substation equipment, integrate BIM data with GIS geographic information, and perform lightweight processing; Real-time collection of multi-source heterogeneous data from substations, including equipment status data, environmental perception data, and video surveillance data; Establish the topological relationship between devices, sensors and events through the graph database and generate a dynamic association graph; Use AI models to integrate and analyze multi-source data, predict equipment degradation trends, and generate optimization decisions; Render multi-source data in layers on a 3D visualization platform to achieve multi-dimensional interactive presentation of device status and decision feedback.

2. The digital twin substation multi-source data visualization and decision support method according to claim 1 is characterized by: The construction of the three-dimensional holographic digital twin model includes: generating a detailed equipment-level model based on a BIM modeling tool and exporting topological relationships through the IFC format; superimposing GIS terrain data and using spatial mosaic technology to match the BIM model with the geographic coordinate system; and lightweight processing of the model, including geometric simplification, texture compression, and format conversion to glTF or FBX format.

3. The digital twin substation multi-source data visualization and decision support method according to claim 1 is characterized by: The multi-source heterogeneous data is collected using edge computing nodes, including support for IEC61850, Modbus and MQTT protocols; data standardization processing includes timestamp alignment, noise filtering and real-time synchronization based on the OPC UA protocol, and is stored in a time series database.

4. The digital twin substation multi-source data visualization and decision support method according to claim 1 is characterized in that: The establishment of the topological association relationship includes: Define node types as devices, sensors, electrical nodes, and events; build physical connections, electrical paths, and event trigger relationships; parse the adjacency matrix based on the single-line diagram to automatically generate physical connection relationships between devices.

5. The digital twin substation multi-source data visualization and decision support method according to claim 1 is characterized in that: The AI ​​model includes: The LSTM neural network predicts the remaining service life of the equipment; the knowledge graph matches the historical failure case library to quantify the failure probability and risk weight; the multi-objective optimization model generates a Pareto optimal solution set, combining equipment life, operating efficiency and maintenance cost to make a trade-off.

6. The digital twin substation multi-source data visualization and decision support method according to claim 1 is characterized in that: The implementation of the layered rendering includes: Infrared thermal imaging data is mapped to the device surface, and a dynamic thermal distribution map is generated through GPU shaders. A particle system is used to visualize partial discharge signals, with the intensity driven by the discharge quantity parameter. Touch screen interaction is supported, and device-related data, including vibration spectra and oil chromatogram historical trends, can be retrieved through X-ray detection.

7. The digital twin substation multi-source data visualization and decision support method according to claim 1 is characterized in that: The generation of the optimization decision is specifically as follows: When the predicted equipment degradation threshold exceeds the limit, maintenance recommendations are automatically pushed and spare parts allocation orders are generated; the transformer operation strategy is adjusted according to real-time load data to balance equipment life and energy efficiency; historical fault scenarios are replayed through the digital twin model to optimize the protection setting.

8. A digital twin substation multi-source data visualization decision-making device, characterized in that: include: Twin modeling module, used to build lightweight 3D models and achieve dynamic data synchronization; Data fusion module, used for real-time collection, standardization and graph database storage of multi-source heterogeneous data; AI decision-making module for equipment status prediction, risk assessment, and strategy optimization; Visual interaction module for multi-layer rendering, touch screen operation and decision feedback.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The storage medium includes a plurality of program codes, and the program codes are used to be loaded and run by a processor to execute the method according to any one of claims 1 to 7.

Images