Multi-source data fusion transformer substation digital twinborn model construction method and system

Abstract

The invention discloses a multi-source data fusion transformer substation digital twinning model construction method and system, and belongs to the technical field of intelligent power grid digital twinning, and the method comprises the steps: carrying out the multi-source data collection to obtain the related data of a transformer substation, and constructing a three-dimensional entity model; deploying a multi-physical field monitoring device, constructing a real-time data acquisition network, and dynamically fusing multi-source data and a three-dimensional entity model; performing feature extraction and weighted fusion on the dynamically fused multi-source data; a multi-physics field coupling analysis model is constructed based on a deep learning network, and dynamic simulation and visualization of multiple physics fields are realized by using fusion feature data; system function modularization is achieved through a micro-service architecture, real-time updating, cross-platform deployment and multi-user interaction of a multi-physics coupling analysis model are supported, and a transformer substation digital twinborn model is obtained. According to the method, data of multiple physical quantities can be fused, multi-physical field coupling analysis is carried out, the overall layout and actual errors are optimized, and the precision and accuracy of the model are improved.

Description

Technical Field

[0001] This invention belongs to the field of digital twin technology for smart grids, specifically relating to a method and system for constructing a digital twin model of a substation by fusing multi-source data. Background Technology

[0002] With the deepening of smart grid construction, substations, as key nodes in the power system, are receiving increasing attention for their operational safety and reliability. In recent years, the power grid sector has prioritized improving sensor quality and reliability, explicitly pointing out that the intensity and distribution patterns of the electromagnetic environment in which sensors operate within substations are not yet fully understood, making it difficult to define the electromagnetic compatibility performance requirements of sensors, thus affecting their operational reliability. Traditional substation modeling methods often rely on a single data source or simplified simulation models, making it difficult to accurately reflect the complex multi-physics coupling characteristics within substations, especially during transient processes such as electromagnetic transients, where model accuracy and real-time performance are significantly insufficient.

[0003] As an improvement, while digital twin technology has seen initial applications in the industrial sector, it still faces challenges in substation scenarios. For example, substations involve data on multiple physical quantities such as electromagnetics, temperature, and vibration, sourced from sources including laser scanning, UAV mapping, high-precision CAD drawings, and electromagnetic sensors. These data vary significantly in format, sampling frequency, and spatiotemporal scale, making data fusion difficult. Traditional modeling methods tend to overlook local details in electromagnetic transient feature extraction, leading to large errors between the model and actual equipment, making it difficult to meet millimeter-level accuracy requirements. Most models rely on offline simulation, failing to achieve real-time synchronous mapping with physical entities, thus hindering dynamic early warning and decision-making. Systems often focus on displaying single physical fields, lacking dynamic simulation and visualization of multi-field coupling (electric, magnetic, temperature, etc.), and missing multi-physics coupling analysis. In summary, current substation modeling methods suffer from high data heterogeneity, making data fusion difficult. Local details are difficult to extract in electromagnetic transient feature extraction, resulting in large errors. Dynamic simulation and analysis of multi-physics coupling are insufficient, and the accuracy and real-time performance of the models require further optimization. Summary of the Invention

[0004] This invention provides a method and system for constructing a digital twin model of a substation by fusing multi-source data. The purpose is to solve the problems in current substation modeling methods, such as the high heterogeneity of data, which makes it difficult to fuse the data, the difficulty in extracting local details in electromagnetic transient feature extraction leading to large errors, and the lack of dynamic simulation and analysis of multi-physics coupling, which result in the need for further optimization of the model's accuracy and real-time performance.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for constructing a digital twin model of a substation based on multi-source data fusion, comprising the following steps: S1. Acquire relevant data about the substation through multi-source data acquisition methods and construct a three-dimensional solid model; S2. Deploy multi-physics monitoring equipment, build a real-time data acquisition network, and use data assimilation technology to achieve dynamic fusion of multi-source data and three-dimensional entity models to obtain dynamically fused multi-source data. S3. Employ a multi-source data fusion network to extract features and perform weighted fusion on the dynamically fused multi-source data to obtain fused feature data. S4. Construct a multi-physics coupling analysis model based on deep learning networks, and use fused feature data to realize the dynamic simulation and visualization of multi-physics. S5. The system functions are modularized through a microservice architecture, supporting real-time updates, cross-platform deployment, and multi-user interaction of the multi-physics coupling analysis model, resulting in a digital twin model of the substation and completing the construction of a digital twin model of the substation with multi-source data fusion. The multi-source data fusion network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. The deep learning network includes a Transformer neural network and a U-Net network. Both the multi-source data fusion network and the deep learning network are pre-trained networks.

[0006] In some implementations, in S1, the multi-source data acquisition methods include laser scanning, UAV mapping, and reverse engineering of high-precision CAD drawings; the acquired substation-related data includes geometric data and three-dimensional material data of substation equipment and architecture.

[0007] In some implementations, in S1, geometric data is acquired through 3D scanning using a laser scanner and a UAV mapping system, 3D material data is captured using multispectral imaging technology, all acquired data is uniformly converted to IFC format, and spatial registration is performed using the WGS84 coordinate system.

[0008] In some implementations, in S2, the multi-physics monitoring device includes an electromagnetic sensor, a temperature sensor, and a flow sensor. The multi-source data collected in real time is preprocessed locally through an edge computing node. The data assimilation technology is implemented through the Kalman filter layer in the multi-source data fusion network to solve the problems of time asynchrony and noise in the sensor data, and to achieve data state estimation and error correction.

[0009] In some implementations, in S3, the encoder-decoder layer employs a three-dimensional convolutional layer, a batch normalization layer, and a pooling / upsampling layer structure; the feature map calculation formula for the encoder-decoder layer is as follows: ; in, For the encoder-decoder layer Layer Each output feature map; It is a linear rectification activation function; For the encoder-decoder layer Layer The bias term corresponding to each output feature map; The index variable for the input feature map; This represents the total number of input feature maps; For the encoder-decoder layer Layer The output feature map corresponds to the first... Filter weights for each input feature map; For the encoder-decoder layer Layer Input feature maps.

[0010] In some implementations, in S3, the adaptive weighted fusion layer dynamically learns the importance of each data source through a sigmoid activation function and calculates the weights according to the following formula: ; ; ; in, These are the weighting coefficients corresponding to the features of the satellite cloud image; These are the weighting coefficients corresponding to the features of the radar echo map; These are the weighting coefficients corresponding to the meteorological factor characteristics; It is the Sigmoid activation function. ; The weights of the convolution kernels corresponding to the satellite cloud image features; The convolution kernel weights are the features corresponding to the radar echo map. The convolution kernel weights are the features of the meteorological factors. , , These are satellite cloud images, radar echo images, and meteorological factor characteristics, respectively. The bias term used in the calculation of satellite cloud image feature weights; The bias term used to calculate the feature weights of the radar echo map; This is the bias term used to calculate the characteristic weights of meteorological factors.

[0011] In some implementations, in S4, the Transformer neural network uses a self-attention mechanism to capture long sequence dependencies, calculated as follows: ; in, This is the output of the self-attention mechanism; This is a query matrix that stores the feature information to be queried. The key matrix; It is a value matrix; To find the product of the key matrix and the transpose of the key matrix; The dimension of the key vector; This is the scaling factor; This is the activation function.

[0012] In some implementations, in S4, the U-Net network fuses encoder and decoder features through skip connections, and the formula for calculating the output segmentation probability map is: ; in, The segmentation probability map output by the U-Net network; This represents the total number of eigenvectors after dimensionality reduction. These are the eigenvectors after dimensionality reduction; For the first The weights and biases of the fully connected layer corresponding to each feature vector; For the first The bias of the fully connected layer corresponding to each feature vector.

[0013] In some implementations, in S5, the microservice architecture breaks down system functions into independent modules such as data acquisition services, fused computing services, and visualization services. It adopts Docker containerization for deployment and supports dynamic scheduling by Kubernetes. The system supports WebGL, DirectX, XML, HTTP, TCP / IP, and Unity3D standard protocols, and uses a microservice architecture to achieve module decoupling and dynamic expansion.

[0014] This invention also provides a substation digital twin model construction system for multi-source data fusion, used to realize the above-mentioned substation digital twin model construction method for multi-source data fusion, including a three-dimensional solid model module, a dynamic fusion module, a feature extraction and weighted fusion module, a multi-physics coupling analysis module, and a digital twin model generation module, wherein: 3D solid model module: used to acquire relevant data of substation through multi-source data acquisition methods and construct a 3D solid model; Dynamic fusion module: used to deploy multi-physics monitoring equipment to build a real-time data acquisition network, and to use data assimilation technology to achieve dynamic fusion of multi-source data and three-dimensional entity models to obtain dynamically fused multi-source data; Feature extraction and weighted fusion module: Used to extract features and perform weighted fusion on dynamically fused multi-source data using a multi-source data fusion network to obtain fused feature data; Multiphysics Coupling Analysis Module: Used to build multiphysics coupling analysis models based on deep learning networks, and to realize dynamic simulation and visualization of multiphysics using fused feature data; Digital twin model generation module: Used to realize modular system functions through microservice architecture, support real-time updates of multi-physics coupling analysis models, cross-platform deployment and multi-user interaction, obtain substation digital twin models, and complete the construction of substation digital twin models with multi-source data fusion; The multi-source data fusion network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. The deep learning network includes a Transformer neural network and a U-Net network. Both the multi-source data fusion network and the deep learning network are pre-trained networks.

[0015] Compared with existing technologies, the present invention provides a method and system for constructing a digital twin model of a substation based on multi-source data fusion, which has the following advantages: This invention presents a method for constructing a digital twin model of a substation based on multi-source data fusion. Through a multi-source data fusion network and a deep learning model, it achieves a significant improvement in the accuracy of the 3D entity model, with an overall layout similarity of no less than 98%, equipment shape error no greater than 5mm, connection component position error no greater than 3mm, and angle error no greater than 3°. Through multi-physics coupling analysis based on Transformer and U-Net networks, it supports data updates every second, enabling real-time synchronous mapping between the digital twin model and the physical entity. This invention overcomes the limitations of a single physical field, achieving dynamic coupling simulation and visualization of electric, magnetic, and temperature fields, effectively and accurately providing comprehensive data support for substation fault early warning. The microservice architecture supports modular expansion, is compatible with multiple industry standard protocols, facilitates integration with existing power system platforms, and has excellent applicability. Attached Figure Description

[0016] The accompanying drawings are provided to further understand the invention and constitute a part of this invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0017] Figure 1 This is a schematic diagram of the overall architecture of Transformer in the substation digital twin model construction method of multi-source data fusion according to the present invention; Figure 2 This is a schematic diagram of the self-attention mechanism calculation process in the substation digital twin model construction method of multi-source data fusion according to the present invention. Figure 3 This is a schematic diagram of the multi-head attention mechanism in the substation digital twin model construction method of multi-source data fusion according to the present invention; Figure 4 This is a flowchart illustrating the system functional modules in the method for constructing a digital twin model of a substation using multi-source data fusion according to the present invention. Figure 5 This is a schematic diagram of the U-Net network encoder-decoder structure in the substation digital twin model construction method of multi-source data fusion according to the present invention; Figure 6 This is a flowchart illustrating a method for constructing a digital twin model of a substation using multi-source data fusion, as described in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to indicate that selected embodiments of the invention are based on the embodiments in this invention. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, elements defined by the phrase "comprising one..." do not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. It should be noted that the apparatus and methods disclosed in the embodiments herein can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functions, and operations of possible implementations of apparatus, methods, and computer program products according to various embodiments herein. In this respect, each block in a flowchart or block diagram may represent a module, program, or part of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the accompanying drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system to perform the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions. In addition, the functional modules in the various embodiments of this article can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0020] This paper proposes a method for constructing a digital twin model of a substation based on multi-source data fusion. By fusing heterogeneous data from multiple sources and introducing a deep learning-driven feature extraction and adaptive weighted fusion mechanism, a high-precision, real-time updated electromagnetic transient digital twin model of a substation can be constructed.

[0021] Based on this, the present invention provides a method for constructing a digital twin model of a substation through multi-source data fusion, comprising the following steps: S1. Acquire relevant data about the substation through multi-source data acquisition methods and construct a three-dimensional solid model; S2. Deploy multi-physics monitoring equipment, build a real-time data acquisition network, and use data assimilation technology to achieve dynamic fusion of multi-source data and three-dimensional entity models to obtain dynamically fused multi-source data. S3. Employ a multi-source data fusion network to extract features and perform weighted fusion on the dynamically fused multi-source data to obtain fused feature data. S4. Construct a multi-physics coupling analysis model based on deep learning networks, and use fused feature data to realize the dynamic simulation and visualization of multi-physics. S5. The system functions are modularized through a microservice architecture, supporting real-time updates, cross-platform deployment, and multi-user interaction of the multi-physics coupling analysis model, resulting in a digital twin model of the substation and completing the construction of a digital twin model of the substation with multi-source data fusion. The multi-source data fusion network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. The deep learning network includes a Transformer neural network and a U-Net network. Both the multi-source data fusion network and the deep learning network are pre-trained networks.

[0022] The method of this invention acquires relevant data from substations through multi-source data acquisition and constructs a three-dimensional entity model, integrating multiple types of key data from the data source; deploys multi-physics monitoring equipment to build a real-time data acquisition network, and uses data assimilation technology to achieve dynamic fusion of multi-source data and the three-dimensional entity model. The real-time acquisition network ensures data timeliness and improves the real-time mapping capability between the model and the physical entity; a multi-source data fusion network including a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer is used to extract features and perform weighted fusion on the dynamically fused data, effectively extracting and weighting key features, and providing high-quality fused data for multi-physics analysis.

[0023] This invention constructs a multiphysics coupling analysis model based on a deep learning network that includes a Transformer neural network and a U-Net network. It utilizes fused feature data to achieve dynamic simulation and visualization of multiphysics, improves the accuracy of the simulation through the characteristics of deep learning, and optimizes the precision of the model. It achieves functional modularization through a microservice architecture, supports real-time model updates, cross-platform deployment, and multi-user interaction, and constructs a digital twin model of a substation that meets the requirements.

[0024] Furthermore, the multi-source data acquisition methods of this invention include laser scanning, UAV mapping, and reverse engineering of high-precision CAD drawings. The acquired data consists of geometric data and three-dimensional material data. The data is uniformly converted to IFC format and registered using the WGS84 coordinate system to ensure higher data source quality and better data consistency for the three-dimensional solid model, thereby improving the geometric accuracy of the model. The multi-physics monitoring equipment of this invention includes electromagnetic sensors, temperature sensors, and flow sensors. Local preprocessing is performed through edge computing nodes, and the data assimilation technology is based on the Kalman filter layer of the multi-source data fusion network, effectively solving the problems of time asynchrony and noise in sensor data and improving the accuracy of dynamic fusion data.

[0025] Furthermore, the encoder-decoder layer of this invention employs a three-dimensional convolutional layer, batch normalization layer, and pooling / upsampling layer structure. The adaptive weighted fusion layer improves the accuracy of fused feature data through the Sigmoid activation function and dynamically learned data source weights. The Transformer neural network self-attention mechanism and the U-Net network output segmentation probability map ensure the accuracy of multi-physics dynamic simulation and the reliability of visualization. The data acquisition service, fusion computing service, and visualization service of this invention's microservice architecture are deployed using Docker containers and scheduled with Kubernetes, and are compatible with WebGL, DirectX, XML, HTTP, TCP / IP, and Unity3D standard protocols. The system has scalability and adaptability, ensuring stable operation of real-time model updates and multi-user interaction functions.

[0026] Specifically, in some practical working conditions, the method of the present invention is carried out according to the following steps: By using laser scanning, UAV mapping, and reverse engineering of high-precision CAD drawings, geometric and three-dimensional material data of substation equipment and architecture are collected to construct a three-dimensional solid model with millimeter-level precision. Deploy multi-physics field monitoring equipment such as electromagnetic sensors, temperature sensors, and flow sensors to build a real-time data acquisition network, and dynamically fuse multi-source data with digital twin models through data assimilation technology; A multi-source data fusion network is used to extract features and perform weighted fusion on the collected data. The network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. Based on Transformer neural network and U-Net network, a multiphysics coupling analysis model is constructed to realize dynamic simulation and thermographic display of electric field, magnetic field and temperature field; The system functions are modularized through a microservice architecture, which supports real-time model updates, cross-platform deployment, and multi-user interaction.

[0027] This invention utilizes multi-source data acquisition and 3D solid modeling to collect geometric and 3D material data of substation equipment and architecture through laser scanning, UAV mapping, and reverse engineering of high-precision CAD drawings, constructing a 3D solid model with millimeter-level precision. The requirements include an overall layout similarity of ≥98%, equipment shape error ≤±5mm, connection component position error ≤±3mm, and angle error ≤±3°. Simultaneously, multi-physics monitoring devices such as electromagnetic sensors, temperature sensors, and flow sensors are deployed to construct a real-time data acquisition network, and multi-source data is dynamically fused with the digital twin model through data assimilation technology.

[0028] Furthermore, the multi-source data fusion network construction of the present invention employs a multi-source data fusion network to perform feature extraction and weighted fusion of the collected data. The network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. The encoder-decoder layer utilizes a three-dimensional convolutional layer, a batch normalization layer, and a pooling / upsampling layer structure.

[0029] The adaptive weighted fusion layer of this invention dynamically learns the importance of each data source through the Sigmoid activation function.

[0030] The multiphysics coupling analysis model of this invention is based on Transformer neural network and U-Net network to construct a multiphysics coupling analysis model; Transformer neural network uses self-attention mechanism to capture long sequence dependencies.

[0031] The U-Net network of this invention fuses encoder and decoder features through skip connections to output a segmentation probability map.

[0032] This invention achieves modular system functionality through a microservice architecture, supporting real-time model updates, cross-platform deployment, and multi-user interaction. The system is compatible with standard protocols such as WebGL, DirectX, XML, HTTP, TCP / IP, and Unity3D, ensuring high availability and scalability.

[0033] The present invention will be further described in detail below through specific embodiments.

[0034] like Figures 1-6 As shown, this invention relates to a method for constructing a digital twin model of a substation based on multi-source data fusion. Step 1: Construct multi-source data acquisition and 3D solid model; High-precision measurement methods are used to acquire geometric and physical data of substation equipment and environment, laying the foundation for digital twin models. The specific implementation is as follows: Using laser scanners (such as terrestrial LiDAR) and UAV mapping systems, 3D scanning of key equipment in the substation, including transformers, circuit breakers, and disconnect switches, is performed with millimeter-level precision, achieving a point cloud density of no less than 1000 points per square meter. Simultaneously, reverse engineering is performed using high-precision CAD drawings to extract structural parameters of the equipment (such as dimensions and the location of connecting components), ensuring that the model's external shape error is ≤ ±5mm and its angular error is ≤ ±3°.

[0035] Multispectral imaging technology captures the surface material properties of devices (such as metal reflectivity and insulating material texture) and integrates them into a 3D model, supporting the realism of physical field simulation.

[0036] All collected data were uniformly converted to IFC (Industrial Basic Class) format and spatially registered using the WGS84 coordinate system to ensure that the overall layout similarity of the model is ≥98%.

[0037] Step 2: Construct a real-time data acquisition network for data assimilation; To achieve dynamic synchronization between the digital twin model and the physical entity, multi-physics monitoring equipment was deployed and a data fusion mechanism was established; Electromagnetic sensors (to measure electric and magnetic field strength), temperature sensors (to monitor equipment temperature rise), and flow sensors (to detect coolant flow) are deployed at key nodes in the substation, with sampling frequencies up to 1kHz to ensure real-time performance.

[0038] A Kalman filter layer is used to perform preliminary fusion of multi-source data, addressing the issues of temporal asynchrony and noise in sensor data. The Kalman filter formula is based on the filtering principle in document 1, achieving state estimation and error correction.

[0039] Local preprocessing is performed via edge computing nodes, reducing the cloud transmission load. Data upload cycle is 5 seconds, supporting dynamic updates of the digital twin model.

[0040] Step 3: Feature extraction and weighted fusion of multi-source data fusion network. Deep learning network is used to deeply fuse heterogeneous data. The feature extraction and fusion mechanism is as follows: The multi-source data fusion network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. The encoder-decoder layer employs a 3D convolutional layer, a batch normalization layer, and a pooling / upsampling layer structure. Its feature map calculation formula is as follows: ; in, For the encoder-decoder layer Layer Each output feature map; This is a linear rectification activation function used to introduce nonlinear characteristics; For the encoder-decoder layer Layer The bias term corresponding to each output feature map; The index variable for the input feature map; This represents the total number of input feature maps; For the encoder-decoder layer Layer The output feature map corresponds to the first... Filter weights for each input feature map; For the encoder-decoder layer Layer Input feature maps.

[0041] An adaptive weighted fusion of geometric data, sensor data, and environmental features (such as temperature fields) is performed, with the weights calculated using the following formula: ; ; ; in, These are the weighting coefficients corresponding to the features of the satellite cloud image; These are the weighting coefficients corresponding to the features of the radar echo map; These are the weighting coefficients corresponding to the meteorological factor characteristics; It is the Sigmoid activation function. Used to map weight values ​​to interval; The weights of the convolution kernels corresponding to the satellite cloud image features; The convolution kernel weights are the features corresponding to the radar echo map. The convolution kernel weights are the features of the meteorological factors. , , These are satellite cloud images, radar echo images, and meteorological factor characteristics, respectively. The bias term used in the calculation of satellite cloud image feature weights; The bias term used to calculate the feature weights of the radar echo map; This is the bias term used to calculate the characteristic weights of meteorological factors.

[0042] Fusion features are generated through fully connected layers to support subsequent physics field analysis. The loss function used is: ; in, This is the loss function value, used to measure the deviation between the model's predictions and the actual results; This represents the true value vector of the physical fields (electric field, magnetic field, temperature field) of the substation; This is the vector of physical field values ​​predicted by the model; This represents the total number of training samples; For the first The model prediction value corresponding to each sample; This is the averaging coefficient, used to normalize the loss value.

[0043] Step 4: Construct a multiphysics coupling analysis model; Based on Transformer neural networks and U-Net networks, dynamic simulation and visualization of electric, magnetic, and temperature fields are achieved: like Figure 1As shown, the Transformer model is used to process time-series sensor data (such as electromagnetic transient signals). The self-attention mechanism calculation formula is: ; in, This is the output of the self-attention mechanism, used to capture long-sequence dependencies in time-series data; This is a query matrix that stores the feature information to be queried. This is the key matrix, used to calculate relevance when matched with the query matrix; It is a value matrix that stores the eigenvalue information corresponding to the key matrix; This is the transpose of the query matrix and the key matrix, used to calculate the relevance between the query and the key. The dimension of the key vector; This is a scaling factor used to avoid The gradient vanishing problem caused by excessively large values; This is the activation function used to normalize the correlation results into a probability distribution.

[0044] U-Net networks are used for spatial field prediction, such as Figure 5 As shown, U-Net employs an encoder-decoder structure, fusing shallow details and deep features through skip connections. The formula for calculating the output segmentation probability map is as follows: ; in, The segmentation probability map output by the U-Net network is used to represent the probability distribution of physical fields (such as temperature fields) in the substation at various locations in space. This represents the total number of eigenvectors after dimensionality reduction. The feature vector after dimensionality reduction integrates the fused features of the encoder and decoder; For the first The weights and biases of the fully connected layer corresponding to each feature vector; For the first The bias of the fully connected layer corresponds to each feature vector; this model simulates the temperature field of the substation as an image segmentation task and outputs a heat map.

[0045] Step 5. Deploy the microservice architecture and system; Modular design supports system scalability and real-time interaction: The system functions are broken down into independent modules such as data acquisition service, fusion computing service, and visualization service, and are deployed using Docker containers, supporting dynamic scheduling by Kubernetes.

[0046] The system supports WebGL, DirectX, XML, HTTP, TCP / IP, and Unity3D standard protocols, and users can interact through a web interface or VR devices.

[0047] The model synchronizes sensor data every 30 seconds and supports incremental learning: when the prediction error exceeds the threshold, the model is automatically retrained.

[0048] Furthermore, this embodiment takes a 500kV substation as an example; Laser scanning covers all equipment within the station, generating a 3D model; 50 electromagnetic sensors and 30 temperature sensors are deployed, with data acquisition latency <100ms.

[0049] The fusion network was trained using three years of historical operation and maintenance data. The Transformer model input window consisted of 24-hour time series data, while the U-Net input consisted of a 256×256 resolution temperature field image.

[0050] The digital twin model predicts temperature field error ≤ ±2°C, and the correlation coefficient between electromagnetic field simulation and actual measurement is ≥ 0.95, meeting the accuracy requirements.

[0051] This invention discloses a method and system for constructing a digital twin model of a substation based on multi-source data fusion. Through multi-source data fusion and a deep learning model, a high-fidelity digital twin of the substation is constructed. The data acquisition phase ensures the geometric accuracy of the model; data assimilation and fusion networks address the consistency problem of heterogeneous data; Transformer and U-Net models respectively handle temporal and spatial features, achieving dynamic mapping of multi-physics fields; and a microservice architecture ensures the system's real-time performance and scalability. When a substation failure occurs, the twin model can simulate the failure evolution, providing technical support for operation and maintenance decisions.

[0052] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Anyone skilled in the art can readily implement the present invention according to the description and above. Any modifications, alterations, or equivalent variations made using the technical content disclosed above are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A method for constructing a digital twin model of a substation through multi-source data fusion, characterized in that, Includes the following steps: S1. Acquire relevant data about the substation through multi-source data acquisition methods and construct a three-dimensional solid model; S2. Deploy multi-physics monitoring equipment, build a real-time data acquisition network, and use data assimilation technology to achieve dynamic fusion of multi-source data and three-dimensional entity models to obtain dynamically fused multi-source data. S3. Employ a multi-source data fusion network to extract features and perform weighted fusion on the dynamically fused multi-source data to obtain fused feature data. S4. Construct a multi-physics coupling analysis model based on deep learning networks, and use fused feature data to realize the dynamic simulation and visualization of multi-physics. S5. The system functions are modularized through a microservice architecture, supporting real-time updates, cross-platform deployment, and multi-user interaction of the multi-physics coupling analysis model, resulting in a digital twin model of the substation and completing the construction of a digital twin model of the substation with multi-source data fusion. The multi-source data fusion network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. The deep learning network includes a Transformer neural network and a U-Net network. Both the multi-source data fusion network and the deep learning network are pre-trained networks.

2. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In S1, the multi-source data acquisition methods include laser scanning, UAV mapping, and high-precision CAD drawing reverse engineering; the acquired substation-related data includes geometric data and three-dimensional material data of substation equipment and architecture.

3. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In S1, geometric data is acquired through 3D scanning using a laser scanner and a UAV mapping system, and 3D material data is captured using multispectral imaging technology. All acquired data is uniformly converted to IFC format and spatially registered using the WGS84 coordinate system.

4. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In S2, the multiphysics monitoring device includes an electromagnetic sensor, a temperature sensor, and a flow sensor. The multi-source data collected in real time is preprocessed locally through an edge computing node. The data assimilation technology is implemented through the Kalman filter layer in the multi-source data fusion network to solve the problems of time asynchrony and noise in the sensor data, and to achieve data state estimation and error correction.

5. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In S3, the encoder-decoder layer employs a three-dimensional convolutional layer, a batch normalization layer, and a pooling / upsampling layer structure; the feature map calculation formula for the encoder-decoder layer is as follows: ； in, For the encoder-decoder layer Layer Each output feature map; It is a linear rectification activation function; For the encoder-decoder layer Layer The bias term corresponding to each output feature map; The index variable for the input feature map; This represents the total number of input feature maps; For the encoder-decoder layer Layer The output feature map corresponds to the first... Filter weights for each input feature map; For the encoder-decoder layer Layer Input feature maps.

6. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In S3, the adaptive weighted fusion layer dynamically learns the importance of each data source through the Sigmoid activation function and calculates the weights according to the following formula: ； ； ； in, These are the weighting coefficients corresponding to the features of the satellite cloud image; These are the weighting coefficients corresponding to the features of the radar echo map; These are the weighting coefficients corresponding to the meteorological factor characteristics; It is the Sigmoid activation function. ; The convolution kernel weights are the features of the satellite cloud image. The convolution kernel weights are the features corresponding to the radar echo map. The convolution kernel weights are the features of the meteorological factors. , , These are satellite cloud images, radar echo images, and meteorological factor characteristics, respectively. The bias term used in the calculation of satellite cloud image feature weights; The bias term used to calculate the feature weights of the radar echo map; This is the bias term used to calculate the characteristic weights of meteorological factors.

7. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In S4, the Transformer neural network uses a self-attention mechanism to capture long sequence dependencies, calculated as follows: ； in, This is the output of the self-attention mechanism; For query matrix; The key matrix; It is a value matrix; To find the product of the key matrix and the transpose of the key matrix; The dimension of the key vector; This is the scaling factor; This is the activation function.

8. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In step S4, the U-Net network fuses encoder and decoder features through skip connections, and the formula for calculating the output segmentation probability map is as follows: ； in, The segmentation probability map output by the U-Net network; This represents the total number of eigenvectors after dimensionality reduction. These are the eigenvectors after dimensionality reduction; For the first The weights and biases of the fully connected layer corresponding to each feature vector; For the first The bias of the fully connected layer corresponding to each feature vector.

9. The method for constructing a digital twin model of a substation based on multi-source data fusion according to claim 1, characterized in that, In the S5, the microservice architecture breaks down the system functions into independent modules: data acquisition service, fused computing service, and visualization service. It adopts Docker containerization deployment and supports dynamic scheduling by Kubernetes. The system supports WebGL, DirectX, XML, HTTP, TCP / IP, and Unity3D standard protocols, and uses the microservice architecture to achieve module decoupling and dynamic expansion.

10. A system for constructing a digital twin model of a substation through multi-source data fusion, characterized in that, The method for constructing a digital twin model of a substation to achieve multi-source data fusion as described in any one of claims 1-9 includes a three-dimensional solid model module, a dynamic fusion module, a feature extraction and weighted fusion module, a multi-physics coupling analysis module, and a digital twin model generation module, wherein: 3D solid model module: used to acquire relevant data of substation through multi-source data acquisition methods and construct a 3D solid model; Dynamic fusion module: used to deploy multi-physics monitoring equipment to build a real-time data acquisition network, and to use data assimilation technology to achieve dynamic fusion of multi-source data and three-dimensional entity models to obtain dynamically fused multi-source data; Feature extraction and weighted fusion module: Used to extract features and perform weighted fusion on dynamically fused multi-source data using a multi-source data fusion network to obtain fused feature data; Multiphysics Coupling Analysis Module: Used to build multiphysics coupling analysis models based on deep learning networks, and to realize dynamic simulation and visualization of multiphysics using fused feature data; Digital twin model generation module: Used to realize modular system functions through microservice architecture, support real-time updates of multi-physics coupling analysis models, cross-platform deployment and multi-user interaction, obtain substation digital twin models, and complete the construction of substation digital twin models with multi-source data fusion; The multi-source data fusion network includes a Kalman filter layer, an encoder-decoder layer, a correction unit, and an adaptive weighted fusion layer. The deep learning network includes a Transformer neural network and a U-Net network. Both the multi-source data fusion network and the deep learning network are pre-trained networks.

Images