Substation structure state dynamic visualization method and system based on digital twinning

Abstract

The application relates to the technical field of dynamic visualization, in particular to a substation structure state dynamic visualization method and system based on digital twinning, which comprises the following steps: extracting a structure stiffness topology graph and member axial deformation stiffness, and predicting expected stress of each node based on wind load data; comparing actual stress with expected stress, screening out stress positive deviation nodes as tracking end points; generating a global state transition probability matrix based on the stiffness, eliminating end point rows and columns to obtain a non-absorption state probability sub-matrix; inverting a difference between a unit matrix and the non-absorption state probability sub-matrix to obtain an expected path frequency matrix, generating a directed acyclic graph based on the expected path frequency matrix, extracting a load transfer connected path through depth-first search, and visualizing the load transfer connected path on a digital twinning large screen. The application solves the problems of compensation stress mispositioning alarm and long time consumption of finite element real-time calculation, and improves real-time tracking and positioning efficiency of hidden loosening faults of a substation.

Description

Technical Field

[0001] This application relates to the field of dynamic visualization technology, specifically to a method and system for dynamic visualization of substation structural status based on digital twins. Background Technology

[0002] The main steel structure of the substation belongs to a highly statically indeterminate spatial grid system. During long-term operation, the alternating effects of environmental wind loads can easily cause bolts at the flange connections of local components to loosen. When the load-bearing capacity of a local component decreases due to loosening, based on the connectivity characteristics of the statically indeterminate structure, the load it originally bore will be transferred to surrounding healthy components along connected branches according to the stiffness ratio of adjacent components, causing the surface stress of the latter to increase. However, existing online monitoring systems based on digital twin platforms mostly adopt a single-point threshold alarm mechanism based on stress values. This mechanism directly triggers alarms on healthy components that compensate for the stress at the fault point, resulting in a significant spatial misalignment between the warning location output on the digital screen and the actual bolt loosening location.

[0003] To trace the true source of the loosening, the conventional approach in this field is to extract the global stiffness matrix containing all nodal degrees of freedom and then use finite element analysis software to perform nonlinear global static iterative calculations. However, substation structural models typically contain tens of thousands of nodes, making such global iterative inversion calculations time-consuming. In real-time rendering scenarios requiring second-level dynamic refreshes for digital twin displays, excessive computational time leads to severe lag in updating the 3D surface state, failing to meet the frame rate requirements of the visualization system. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for dynamic visualization of substation structural status based on digital twins. The specific technical solution adopted is as follows: In a first aspect, embodiments of this application provide a method for dynamic visualization of the structural status of a substation based on digital twins, the method comprising the following steps: Read the structural stiffness topology diagram from the 3D model of the substation and extract the axial deformation stiffness of each steel component; based on the wind speed and wind direction data in the environment, obtain the expected stress of each node in the structural stiffness topology diagram under the current environment. By comparing the actual stress at the node with the expected stress, nodes with positive stress deviation and nodes with negative stress deviation are selected as the tracking endpoints. Based on the axial deformation stiffness of the steel components between nodes, a global state transition probability matrix describing the topological transition correlation is generated; rows and columns belonging to nodes with positive stress deviation are removed from the global state transition probability matrix to obtain the non-absorbing state probability submatrix. The expected path number matrix is ​​obtained by subtracting the non-absorbing state probability submatrix from the identity matrix and performing a single inversion operation. A directed acyclic graph is generated based on the gradient direction formed by the values ​​in the expected path number matrix. A depth-first search is used to extract the load transfer spatial connectivity path between the stress negative deviation node as the direct connection start point and the stress positive deviation node as the end point. The digital twin screen is then used to visualize the load transfer spatial connectivity path.

[0005] Preferably, the process of constructing the structural stiffness topology diagram is as follows: Extract all steel components and their connection endpoints from the 3D model, treat the connection endpoints as nodes in the undirected graph, treat the steel components as edges in the undirected graph, and denote the resulting undirected graph as the structural stiffness topology graph.

[0006] Preferably, the process for extracting the axial deformation stiffness of each steel component is as follows: Read the cross-sectional parameters of the steel components in the three-dimensional model of the substation. The cross-sectional parameters of the steel components include the cross-sectional area, the elastic modulus of the material, and the length of the steel components. Divide the product of the cross-sectional area and the elastic modulus of the material by the length of the steel components to obtain the axial deformation stiffness of the steel components.

[0007] Preferably, the process of selecting the stress-positive deviation node as the tracking endpoint is as follows: For each node in the structural stiffness topology diagram, the difference between the actual stress at all times within a preset time period before the current time and its respective expected stress is used to form a stress difference sequence. If there are a preset number of consecutive elements in the stress difference sequence of the current node that are all greater than the preset background noise tolerance threshold, then the current node is defined as a stress positive deviation node. If there are a preset number of consecutive elements in the stress difference sequence of the current node that are all less than the negative number of the preset background noise tolerance threshold, then the current node is defined as a stress negative deviation node.

[0008] Preferably, generating the global state transition probability matrix describing topological transition correlation includes: Construction dimension Global state transition probability matrix ,in, The number of nodes in the structural stiffness topology diagram is represented by the number of nodes in the structural stiffness topology diagram. In the structural stiffness topology diagram, the state transition probability of a node flowing to its directly connected node is set to the value of the corresponding element in the global state transition probability matrix.

[0009] Preferably, the method for calculating the state transition probability from a node to its directly connected nodes is as follows: Set the state transition probability of all stresses that are positively deviating from the node to 1, and set the state transition probability of all stresses that are negatively deviating from the node to 0. For all nodes other than the nodes with positive and negative stress deviations, calculate the sum of the axial deformation stiffness of the steel components between all directly connected nodes of each node, and denot it as the total connected stiffness of each node; take the proportion of the axial deformation stiffness of the steel components between each node and any of its directly connected nodes in the total connected stiffness as the state transition probability of each node flowing to any of its directly connected nodes.

[0010] Preferably, the process of generating the directed acyclic graph is as follows: Extract the row vectors of the rows where the stress negative deviation nodes are located from the expected path number matrix, and use the column indexes of the elements with values ​​greater than 0 in the row vectors as the corresponding nodes as transit nodes to construct an undirected topology subnet. In an undirected topological subnet, for any two directly interconnected nodes, the values ​​of the corresponding values ​​of the two nodes in the expected number of traversals matrix are compared. The connection direction of the edge between the two nodes is defined as the node with the larger value pointing to the node with the smaller value, so that the undirected topological subnet is transformed into a directed acyclic graph without closed loops.

[0011] Preferably, the step of extracting the load transfer spatial connectivity path directly connecting the start and end points using depth-first search includes: In a directed acyclic graph, the search starts at nodes with negative stress deviation and ends at nodes with positive stress deviation. A depth-first search algorithm is executed to eliminate isolated edges and invalid connected edges in the directed acyclic graph that cannot reach the endpoint, thereby generating a connected path for load transfer space.

[0012] Preferably, the driving digital twin screen visualizes the spatial connectivity path of load transfer, including: The load transfer spatial connectivity path is mapped to the spatial coordinate system of the substation's three-dimensional model and displayed in a dynamically highlighted form. Secondly, embodiments of this application also provide a substation structural state dynamic visualization system based on digital twins, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described substation structural state dynamic visualization methods based on digital twins.

[0013] This application has at least the following beneficial effects: This application reduces the three-dimensional solid steel frame to a weighted topological graph and combines it with a pre-trained neural network to realize the second-level deduction of theoretical force, which effectively avoids the high time consumption problem of traditional finite element online calculation. It lays the core data foundation for subsequent real-time comparison of abnormal node force states and accurate tracking of hidden load transfer trajectories, taking into account both physical and mechanical mechanisms and real-time rendering frame rate. Furthermore, this application effectively filters out false alarms caused by transient disturbances of gusts and hardware noise by strictly aligning and comparing the measured stress with the dynamic expected benchmark, combined with the "3σ rule" tolerance threshold and multi-frame temporal continuity verification. It accurately extracts the stress negative deviation node that represents the starting point of load loss and the stress positive deviation node that represents the ending point of stress saturation compensation, thus accurately locking the source and sink boundaries of physical tracing for the subsequent pathfinding model. Furthermore, this application accurately reconstructs the microscopic transmission network of "stiffness-dominated distribution" of loads among healthy components by transforming absolute mechanical stiffness into relative probability transfer scalars; at the same time, it uses matrix block recombination technology to physically isolate the compensatory alarm endpoint of stress saturation and extracts pure non-absorbing state probability submatrices, which not only significantly reduces the data scale and computing power overhead of subsequent inversion calculations, but also transforms the complex nonlinear mechanical redistribution problem into a Markov chain convergence model that can be directly solved algebraically, laying a mathematical foundation for the subsequent extraction of hidden danger source tracing paths; Ultimately, this application compresses the graph theory pathfinding process, which originally required countless recursions, into an algebraic solution through a single matrix inversion operation. At the same time, it uses the numerical gradient of the expected number of traversals to force the establishment of a unidirectional potential energy difference, effectively eliminating the path divergence and closed-loop deadlock problems in complex ring topologies. Finally, through depth-first search, it accurately extracts the connected path directly connecting the source of the hidden danger and the compensation endpoint, and presents it intuitively on the digital twin screen in a dynamic highlight form. This achieves second-level accurate extraction and intuitive visualization of the hidden danger source path, improving the real-time tracking and location efficiency of substation hidden loosening faults. Attached Figure Description

[0014] Figure 1 A flowchart illustrating the steps of a digital twin-based dynamic visualization method for substation structural status, provided in one embodiment of this application. Figure 2 This is a schematic diagram of a node classification process provided in one embodiment of this application. Detailed Implementation

[0015] The following description, in conjunction with the accompanying drawings, details the specific scheme of the dynamic visualization method and system for substation structural status based on digital twins provided in this application.

[0016] Please see Figure 1The diagram illustrates a flowchart of a method for dynamic visualization of substation structural status based on digital twins, according to an embodiment of this application. The method includes the following steps: Step S1: Read the structural stiffness topology diagram from the three-dimensional model of the substation and extract the axial deformation stiffness of each steel component; based on the wind speed and wind direction data in the environment, obtain the expected stress of each node in the structural stiffness topology diagram under the current environment.

[0017] In the actual operation of substations, the massive outdoor steel frame is subjected to the dynamic impact of wind loads over a long period. When a steel component (such as a flange connection node) becomes loose, the mechanical load originally borne by that component does not disappear into thin air. Instead, it follows the laws of mechanics, "searching" for new support paths along the connected steel frame. This is the stress redistribution phenomenon in structural mechanics. In this physical process, the surrounding healthy steel components spontaneously "shoulder" the extra load transferred to the loose steel component. However, this load distribution is not an equal distribution, but strictly follows the physical law of "the stronger bears more": the greater the stiffness and the thicker and stronger the steel component, the stronger its ability to resist axial deformation, and naturally, the larger the proportion of the transferred load it receives; conversely, the proportion received by slender and weak steel components is smaller. Therefore, to accurately track the transfer trajectory of this hidden load, the primary prerequisite is to understand the "skeleton" and "physical parameters" of the entire substation steel frame. At the same time, since wind load changes dynamically over time, to determine whether a steel component has become loose, one cannot only look at the absolute stress. It is necessary to know how much force the steel component "should" bear under the current wind force, and use this as a dynamic criterion for judging whether the stress is abnormal.

[0018] Based on the above analysis, this embodiment adopts a dual-parallel data preparation strategy of "offline graph weight solidification" and "online wind load prediction". Specifically, it reads the structural stiffness topology map from the substation's 3D model and extracts the axial deformation stiffness of each steel component; based on wind speed and direction data in the environment, it obtains the expected stress of each node in the structural stiffness topology map under the current environment. The specific process is as follows: S101: Read the structural stiffness topology diagram from the three-dimensional model of the substation and extract the axial deformation stiffness of each steel component.

[0019] First, read the 3D model of the substation, extract all steel components from the 3D model, and identify the connection endpoints connecting each steel component. Assume the total number of extracted connection endpoints is... Number all connection endpoints sequentially, and set the number range to 1. to The above connection endpoints are taken as nodes in the undirected graph, and the steel components are taken as edges in the undirected graph. The resulting undirected graph is denoted as the structural stiffness topology graph.

[0020] Based on this structural stiffness topology diagram, traverse each edge in the structural stiffness topology diagram. Suppose that the current traversal is to a steel member connecting node numbered m and node numbered n, query the data parameter table of the three-dimensional model in reverse. Through the query operation, read the cross-sectional area, material elastic modulus and length of the corresponding steel member.

[0021] Since steel components possess the ability to resist deformation and share loads regardless of whether they are under compression (generating negative stress) or tension (generating normal stress), the specific calculation formula is as follows: ; In the formula, In the structural stiffness topology diagram, node m and node... Axial deformation stiffness of the steel components between them; In the structural stiffness topology diagram, node m and node... The cross-sectional area of ​​the steel components between them; In the structural stiffness topology diagram, node m and node... The material elastic modulus constant of the steel components between them; In the structural stiffness topology diagram, node m and node... The length of the steel components between them.

[0022] Axial deformation stiffness, as understood from the perspective of structural stiffness, characterizes the ability of a steel member to resist elastic deformation under axial stress. It reflects the physical potential of the member to share transferred loads in the mechanical transmission network. A larger axial deformation stiffness indicates a more robust and sturdy steel member or higher material stiffness, making it less prone to deformation during stress redistribution. Therefore, it can bear a larger proportion of transferred loads and will receive a higher flow weight in the subsequent graph-based pathfinding model, prompting the load to preferentially be transmitted along this high-stiffness path. Conversely, a smaller axial deformation stiffness indicates a more slender and weak steel member, with a poorer resistance to deformation. It shares a lower proportion of loads during transfer, and the probability of the load flowing to that node in topological pathfinding decreases accordingly. This objectively reflects the mechanical law of "stiffness-dominated distribution" at the mathematical model level.

[0023] S102: Based on wind speed and direction data in the environment, obtain the expected stress of each node in the structural stiffness topology diagram under the current environment.

[0024] Since online iterative calculations using conventional finite element analysis software are extremely time-consuming and cannot meet the real-time rendering requirements of 3D large screens, an offline method is adopted to generate stress samples and pre-train neural network models to establish a nonlinear mapping relationship between wind and meteorological conditions and structural stress.

[0025] First, meteorological statistics for the substation's location are retrieved to obtain the lower and upper limits of extreme wind speeds that the substation may encounter throughout the year, as well as the omnidirectional wind direction range (0 to 360 degrees). Then, the area between the lower and upper limits of extreme wind speeds is divided into multiple consecutive wind speed intervals (e.g., each interval is 2 m / s). Simultaneously, the omnidirectional wind direction range is divided into multiple consecutive wind direction angle intervals (e.g., each interval is 15 degrees).

[0026] Furthermore, a random sampling algorithm is employed to randomly select a specific wind speed value within each wind speed range and a specific wind direction angle value within each wind direction angle range. These selected wind speed and wind direction angle values ​​are then combined to generate a set of combined external load parameters representing specific external wind environment conditions. By repeatedly executing the above random sampling and combination operations, several sets of combined external load parameters covering various possible wind environments under normal substation operating conditions are generated.

[0027] Furthermore, the generated combined external load parameters are input one by one into the finite element analysis solver to perform forward static calculations. Through these calculations, the stress and force results of each node in the 3D model under different combined external load parameters are obtained under a healthy state without any loose flange connections. The stress and force results of the nodes are extracted and used as label data. These labels are then paired and associated one-to-one with the corresponding combined external load parameters used as input features to generate an offline dataset for deep learning training.

[0028] A pre-defined stress prediction graph neural network model is invoked, specifically a regression model combining a graph convolutional network (GCN) and a multilayer perceptron (MLP). The offline dataset is divided into a training set and a validation set in an 8:2 ratio. The combined external load parameters (i.e., wind speed and wind direction angle values) and the structural stiffness topology graph in the training set are used as inputs to the regression model combining the graph convolutional network (GCN) and the multilayer perceptron (MLP). After internal calculation, the stress prediction values ​​of each node are output. The stress results of the nodes in the offline dataset are used as the label data for the desired fit. The model is then run through multiple rounds using a backpropagation algorithm and an optimizer. Iterative training involves continuously calculating the mean square error (MSE) between the predicted stress values ​​output by the model and the stress results during training. When the MSE converges to a set acceptable error, training stops, and the connection weight parameters of the model are fixed. Simultaneously, the fixed model is validated using the validation set to confirm that the model still possesses accurate generalization prediction capabilities on data not used in training. This completes the training of the stress prediction neural network. In this embodiment, the acceptable error is 0.0001. In practical applications, as other implementation methods, implementers can set the acceptable error according to specific circumstances. This embodiment does not impose any special limitations.

[0029] The process of training the regression model combining Graph Convolutional Network (GCN) and Multilayer Perceptron (MLP) is a well-known technique and will not be elaborated further. The current wind speed and direction are used as inputs to the stress prediction neural network, which outputs the expected stress of each node in the current structural stiffness topology.

[0030] Thus, this embodiment effectively avoids the high time consumption problem of traditional finite element online calculation by reducing the three-dimensional solid steel frame to a weighted topological graph and combining it with a pre-trained neural network to realize the theoretical force in seconds. This lays the core data foundation for subsequent real-time comparison of abnormal node force states and accurate tracking of implicit load transfer trajectories, taking into account both physical and mechanical mechanisms and real-time rendering frame rate.

[0031] Step S2: Compare the actual stress at the node with the expected stress, and select the nodes with positive stress deviation and nodes with negative stress deviation as the tracking endpoints.

[0032] In actual substation operation, when a bolt loosens or a connection fails at a point in the steel frame, the mechanical load originally borne by that component does not disappear. Instead, it "searches" for new support paths along the connected steel frame, which physically manifests as load transfer and stress redistribution. In this implicit mechanical process, on the one hand, the loosened component, due to the loss of its original connection stiffness, experiences a significant decrease in its actual load and exits the main force transmission path. On the other hand, healthy components originally connected to or adjacent to the loosened component are forced to "bear" the transferred extra load to maintain the structural balance of the entire steel frame, leading to a surge in actual stress. These components become "victims" of structural compensation and are the first to trigger overload alarms. Therefore, to accurately track this invisible load transfer trajectory in three-dimensional space, the primary prerequisite is to identify the components playing different physical roles: that is, by comparing the actual stress measured by on-site sensors with the theoretical stress that the component "should" bear under the current wind load, the "source" of the abnormal decrease in stress and the "compensation endpoint" of the abnormal surge in stress can be identified.

[0033] Based on the above analysis, this embodiment compares the actual stress at the node with the expected stress to select the node with the positive stress deviation as the tracking endpoint. The specific process is as follows: The actual stress data of all nodes in the structural stiffness topology diagram corresponding to the substation is read. Specifically, each node is pre-embedded with a fiber optic stress sensor, which can directly collect the deformation of the steel components and convert and output the stress data of the node at the hardware level. The sampling frequency of the actual stress data is set to 10Hz because the frequency of wind load changes at the substation site and the response bandwidth of the strain sensor itself are usually in the several hertz range. The 10Hz sampling rate satisfies the Nyquist sampling theorem to avoid high-frequency mixing and can accurately capture the transient stress fluctuations of the structure caused by gusts. It can also control the amount of data per second within a reasonable range to ensure the second-level real-time rendering frame rate of the digital twin screen. In practical applications, as other implementation methods, the sampling frequency can be other values. If the site is in a typhoon-prone area or a high-frequency fiber optic stress sensor is deployed, the implementer can increase the sampling frequency to 50Hz or 100Hz to obtain a more detailed stress curve, and the number of data frames in the corresponding sliding window should also be increased proportionally. Conversely, if the hardware computing power is limited or only macroscopic slowly varying stress is concerned, it can be lowered to 1Hz to 5Hz. This embodiment does not impose any special restrictions on this.

[0034] For each node in the structural stiffness topology diagram, the difference between the actual stress at all times within a preset time period before the current time and its respective expected stress is used to form a stress difference sequence. In the structural stiffness topology diagram, if there are a preset number of consecutive elements in the stress difference sequence of the current node that are all greater than the preset background noise tolerance threshold, then the current node is defined as a stress positive deviation node; at the same time, if there are a preset number of consecutive elements in the stress difference sequence of the current node that are all less than the opposite number of the preset background noise tolerance threshold, then the current node is defined as a stress negative deviation node.

[0035] Preferably, the node classification process diagram provided in this embodiment is as follows: Figure 2 As shown.

[0036] It should be noted that in this embodiment, the preset duration is 2 seconds and the preset quantity is 5. The preset duration of 2 seconds is set to construct an appropriate time interception boundary, which can completely isolate the historical failure data caused by the transient change of wind load two seconds ago, so as to avoid its interference with the current state judgment, and ensure that the system has enough continuous historical frames for retrieval within the most recent 2-second time window. The preset quantity of 5 is set to force the node force deviation phenomenon to exist continuously for at least half a second, so as to filter out the occasional single-point jump caused by the instantaneous blowing of gusts or electromagnetic pulses. The above values ​​can be other values. The implementer can make adaptive adjustments according to the meteorological turbulence characteristics of the substation location and the system hardware response delay. This embodiment does not impose any special restrictions.

[0037] It should be understood that stress negative deviation nodes and stress positive deviation nodes are used to characterize two extreme physical roles in the structural stress redistribution process. Specifically, stress negative deviation nodes characterize steel components that have lost their original load-bearing capacity due to flange connection loosening or failure; their actual stress is significantly lower than the theoretically expected value, and they act as the source and starting point for load loss in the mechanical transmission network. Stress positive deviation nodes, on the other hand, characterize surrounding healthy steel components that are forced to compensate for the stress of the potential problem source; their actual stress spikes abnormally and continuously exceeds the theoretically expected value, reaching an overload alarm state of stress saturation. In the mechanical transmission network, they act as the physical endpoint for intercepting and absorbing transferred loads. Together, they define the source and sink boundaries of the load transfer topology deduction.

[0038] It should be noted that the process for setting the preset background noise tolerance threshold in this embodiment is as follows: In this embodiment, a preset background noise tolerance threshold is used. The expression is: In the formula, This represents the standard deviation of the environmental noise floor in the factory documentation of the fiber Bragg grating stress sensor.

[0039] It should be noted that the reason for using the preset background noise tolerance threshold is... The expression is set because the complex electromagnetic environment and sensor acquisition circuits at the substation site inevitably generate random thermal noise that follows a Gaussian distribution. According to the "3σ rule" in statistics, 99.7% of random noise fluctuations under a normal distribution will fall within three standard deviations. Setting this threshold to three times the standard deviation of the noise floor is equivalent to defining an absolutely safe tolerance envelope near zero. Any stress difference fluctuation with an absolute value less than this threshold is considered normal electronic noise or minor environmental disturbances and is absorbed and shielded by the system. Only when the stress difference exceeds this statistical safety boundary is it confirmed that a substantial structural stress deviation has occurred. This confidence coefficient can be flexibly adjusted according to the accuracy level of the on-site sensors. If an ultra-low noise high-precision sensor is used, the coefficient can be appropriately lowered to improve sensitivity. This embodiment does not impose any special restrictions on this.

[0040] Thus, this embodiment effectively filters out false alarms caused by transient disturbances of gusts and hardware noise by strictly aligning and comparing the measured stress with the dynamic expected benchmark, combined with the "3σ rule" tolerance threshold and multi-frame temporal continuity verification. It accurately extracts the stress negative deviation node that represents the starting point of load loss and the stress positive deviation node that represents the ending point of stress saturation compensation, so as to accurately lock the source and sink boundaries of physical tracing for the subsequent pathfinding model.

[0041] Step S3: Generate a global state transition probability matrix describing the topological transition correlation based on the axial deformation stiffness of the steel components between nodes; remove rows and columns belonging to nodes with positive stress deviation from the global state transition probability matrix to obtain the non-absorbing state probability submatrix.

[0042] The compensatory alarm node, which has been identified as having a positive stress deviation, has actually intercepted and absorbed the transferred additional load and is in a "terminal state" of stress saturation. After the load flows to this type of node, it no longer continues to be transmitted downstream. Therefore, to accurately deduce this invisible load transfer trajectory, it is necessary to quantify the transmission tendency between nodes based on the stiffness ratio of each connected steel component and set the compensatory alarm node as the physical endpoint of the load transfer.

[0043] Based on the above analysis, this embodiment generates a global state transition probability matrix describing the topological transition correlation according to the axial deformation stiffness of the steel components between nodes; after removing rows and columns belonging to nodes with positive stress deviation from the global state transition probability matrix, a non-absorbing state probability submatrix is ​​obtained, as detailed below: In this embodiment, the total number of all nodes contained in the structural stiffness topology graph is first read. And the axial deformation stiffness bound to each edge; the construction dimension is Global state transition probability matrix In this matrix, the rows and columns of the global state transition probability matrix represent the node numbers in the structural stiffness topology diagram. In the structural stiffness topology diagram, the state transition probability of a node flowing to its directly connected nodes is set to the value of the corresponding element in the global state transition probability matrix. Specifically, in the structural stiffness topology diagram, the state transition probability of all nodes with positive stress deviation is set to 1. For all nodes other than those with positive and negative stress deviation, the sum of the axial deformation stiffness of the steel components between all directly connected nodes of each node is calculated and denoted as the total connected stiffness of each node. The proportion of the axial deformation stiffness of the steel components between each node and any of its directly connected nodes in the total connected stiffness is used as the state transition probability of each node flowing to any of its directly connected nodes.

[0044] Based on the above calculation process of state transition probability, it can be understood that the absolute mechanical stiffness parameter is transformed into a relative probability scalar with a value between 0 and 1. In essence, this quantifies the tendency of a node to distribute and transfer loads to its surrounding adjacent nodes. Since nodes with positive stress deviation are set as absorption states (rotation probability of 1, outflow probability of 0), it indicates that the loads they have intercepted are no longer transmitted outward; while nodes with negative stress deviation are set as outflow sources (probability of 0), indicating that they only serve as the starting point of load loss and do not participate in the transfer and distribution; for nodes outside of these two states, the probability of their loads flowing to any adjacent node depends entirely on the proportion of the stiffness of the steel components between them to the total external stiffness of the node. This process faithfully reproduces the physical load redistribution law of "the greater the stiffness, the more loads are distributed" in statically indeterminate structures at the mathematical level, laying the probability transition foundation for subsequent pathfinding deduction.

[0045] Furthermore, in this embodiment, by removing rows and columns belonging to nodes with positive stress deviations from the global state transition probability matrix, a non-absorbing state probability submatrix is ​​obtained, specifically: To prevent singular matrices from causing algorithm deadlock errors, explicit boundary checks are performed to determine the set of nodes whose stress is positively offset. The set of nodes that are negatively offset from the stress direction Whether both are not empty, subsequent matrix block operations are triggered only when both are not empty (i.e., there is at least one absorbing state endpoint and one tracking start point in the graph network). If any set is empty, it is determined that the current connectivity deduction conditions are not met, the matrix memory is released and the system waits for the next data acquisition.

[0046] When the above triggering conditions are met, the global state transition probability matrix is... Perform matrix standard form block recombination, and through row and column reordering, remove the rows and columns that belong to the stress positive deviation nodes in the global state transition probability matrix, and use the new matrix composed of the remaining rows and columns as the non-absorbing state probability submatrix Q.

[0047] Based on the non-absorbing state probability submatrix, it can be understood that the non-absorbing state probability submatrix is ​​essentially a locally connected graph that physically isolates the stress-positive deviation nodes (i.e., the absorbing state) in the stress-saturated final state from the global network, retaining only the transfer relationships between all unsaturated healthy nodes. By eliminating the rows and columns containing the stress-positive deviation nodes, the non-absorbing state probability submatrix clearly describes all potential microscopic paths of the transferred load during its splitting, convergence, and movement within the intermediate healthy component group before reaching the alarm endpoint. This matrix dimensionality reduction operation not only removes redundant information about the final state of load transmission, significantly reducing the data scale and computation time of subsequent matrix inversion operations, but also ensures that the inversion operation targets a pure "transfer network," thus accurately resolving the ultimate convergence state of the load flow within the healthy topology.

[0048] Thus, this embodiment accurately reconstructs the microscopic transmission network of "stiffness-dominated distribution" of loads among healthy components by transforming absolute mechanical stiffness into a relative probability transfer scalar. At the same time, it uses matrix block recombination technology to physically isolate the compensatory alarm endpoint of stress saturation and extracts a pure non-absorbing state probability submatrix. This not only significantly reduces the data scale and computational cost of subsequent inversion calculations, but also transforms the complex nonlinear mechanical redistribution problem into a Markov chain convergence model that can be directly solved algebraically, laying a mathematical foundation for the subsequent extraction of hidden danger tracing paths.

[0049] Step S4: Subtract the non-absorbed state probability submatrix from the identity matrix and perform a single inversion operation to obtain the expected path number matrix; generate a directed acyclic graph based on the gradient direction formed by the values ​​in the expected path number matrix, and use depth-first search to extract the load transfer spatial connectivity path between the stress negative deviation node as the direct connection start point and the stress positive deviation node as the end point, and drive the digital twin screen to visualize the load transfer spatial connectivity path.

[0050] Because the steel structure of the substation has a large number of symmetrical or ring connections, direct use for path finding would lead to path divergence. It is necessary to establish a "potential energy difference" based on the physical magnitude of this flow depth, and to force the load to flow only from the node with the greater affected depth to the node with the smaller depth. This logically cuts off all possible return branches that could form a closed loop, transforming the complex undirected network into a directed graph with a unique direction, and thus accurately identifying the force transmission channel from the source of the hidden danger to the end point of the compensation.

[0051] Based on the above analysis, this embodiment uses the identity matrix to subtract the non-absorption state probability submatrix and performs a single inversion operation to obtain the expected path frequency matrix; a directed acyclic graph is generated based on the gradient direction formed by the values ​​in the expected path frequency matrix, and a depth-first search is used to extract the load transfer spatial connectivity path directly connecting the start and end points, driving the digital twin screen to visualize the load transfer spatial connectivity path, specifically: In this embodiment, firstly, the non-absorption state probability submatrix is ​​subtracted from the identity matrix and a single inversion operation is performed to obtain the expected path count matrix. For ease of understanding, the calculation expression is given below: ; In the formula, Represents the expected number of paths matrix; Represents the identity matrix; Represents the probability submatrix of the non-absorbing state; This represents the inverse operation of a matrix.

[0052] Furthermore, in this embodiment, a directed acyclic graph is generated based on the gradient direction formed by the values ​​in the desired path count matrix. Specifically: Extract the row vectors of the rows where the stress negative deviation nodes are located from the expected path number matrix. Take the column index of the elements with values ​​greater than 0 in the row vector as the corresponding nodes as transit nodes. Construct an undirected topology subnet using only these candidate transit nodes and their interconnections. Furthermore, in the undirected topological subnet, for any two directly interconnected nodes, the values ​​of the corresponding values ​​of the two nodes in the expected path count matrix are compared, and the connection direction of the edge between the two nodes is specified to be from the node with the larger value to the node with the smaller value, so that the undirected topological subnet is transformed into a directed acyclic graph without closed loops.

[0053] Furthermore, this embodiment is based on a directed acyclic graph and uses depth-first search to extract the load transfer spatial connectivity path directly connecting the start and end points, driving the digital twin screen to visualize the load transfer spatial connectivity path. The specific process is as follows: In a directed acyclic graph, the search starts at nodes with negative stress deviation and ends at nodes with positive stress deviation. A depth-first search algorithm is executed to eliminate isolated edges and invalid connected edges in the directed acyclic graph that cannot reach the endpoint, thereby generating a connected path for load transfer space. The operating principle of the depth-first search algorithm is a well-known technique and will not be elaborated further.

[0054] To clarify, an "isolated edge" refers to a dead-end branch in a directed acyclic graph that is completely disconnected from the source propagation direction of the starting point or is entirely outside the main topology, making it impossible to receive any transferred loads. An "invalid connected edge," on the other hand, refers to a branch that, although it can be traversed forward from the starting point, ultimately points to an intermediate node with no subsequent path (i.e., a dead end), preventing the load from reaching the compensation alarm node. The depth-first search algorithm eliminates these two types of edges, retaining only the set of valid edges that can connect the beginning and end, thereby refining the load transfer connection path that directly connects the hidden danger point to the endpoint.

[0055] Finally, the load transfer spatial connectivity path is mapped to the spatial coordinate system of the substation's 3D model and displayed in a dynamically highlighted form.

[0056] Thus, this embodiment compresses the graph theory pathfinding process, which originally required countless recursions, into an algebraic solution through a single matrix inversion operation. At the same time, it uses the numerical gradient of the expected number of traversals to forcibly establish a unidirectional potential energy difference, effectively eliminating the path divergence and closed-loop deadlock problems in complex ring topologies. Finally, through depth-first search, it accurately extracts the connected path directly linking the source of the hidden danger to the compensation endpoint, and presents it intuitively on the digital twin screen in a dynamic highlight form. This achieves second-level accurate extraction and intuitive visualization of the hidden danger source path, improving the real-time tracking and location efficiency of concealed loosening faults in substations.

[0057] Based on the same inventive concept as the above methods, this application also provides a dynamic visualization system for substation structural status based on digital twins, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described methods for dynamic visualization of substation structural status based on digital twins.

Claims

1. A method for dynamic visualization of substation structural status based on digital twins, characterized in that, The method includes the following steps: Read the structural stiffness topology diagram from the 3D model of the substation and extract the axial deformation stiffness of each steel component; based on the wind speed and wind direction data in the environment, obtain the expected stress of each node in the structural stiffness topology diagram under the current environment. By comparing the actual stress at the node with the expected stress, nodes with positive stress deviation and nodes with negative stress deviation are selected as the tracking endpoints. Based on the axial deformation stiffness of the steel components between nodes, a global state transition probability matrix describing the topological transition correlation is generated; rows and columns belonging to nodes with positive stress deviation are removed from the global state transition probability matrix to obtain the non-absorbing state probability submatrix. The expected path number matrix is ​​obtained by subtracting the non-absorbing state probability submatrix from the identity matrix and performing a single inversion operation. A directed acyclic graph is generated based on the gradient direction formed by the values ​​in the expected path number matrix. A depth-first search is used to extract the load transfer spatial connectivity path between the stress negative deviation node as the direct connection start point and the stress positive deviation node as the end point. The digital twin screen is then used to visualize the load transfer spatial connectivity path.

2. The method for dynamic visualization of substation structural status based on digital twin as described in claim 1, characterized in that, The process of constructing the structural stiffness topology diagram is as follows: Extract all steel components and their connection endpoints from the 3D model, treat the connection endpoints as nodes in the undirected graph, treat the steel components as edges in the undirected graph, and denote the resulting undirected graph as the structural stiffness topology graph.

3. The method for dynamic visualization of substation structural status based on digital twin as described in claim 1, characterized in that, The process for extracting the axial deformation stiffness of each steel component is as follows: Read the cross-sectional parameters of the steel components in the three-dimensional model of the substation. The cross-sectional parameters of the steel components include the cross-sectional area, the elastic modulus of the material, and the length of the steel components. Divide the product of the cross-sectional area and the elastic modulus of the material by the length of the steel components to obtain the axial deformation stiffness of the steel components.

4. The method for dynamic visualization of substation structural status based on digital twin as described in claim 2, characterized in that, The process of selecting the stress-positive deviation nodes as the tracking endpoint is as follows: For each node in the structural stiffness topology diagram, the difference between the actual stress at all times within a preset time period before the current time and its respective expected stress is used to form a stress difference sequence. If there are a preset number of consecutive elements in the stress difference sequence of the current node that are all greater than the preset background noise tolerance threshold, then the current node is defined as a stress positive deviation node. If there are a preset number of consecutive elements in the stress difference sequence of the current node that are all less than the negative number of the preset background noise tolerance threshold, then the current node is defined as a stress negative deviation node.

5. The method for dynamic visualization of substation structural status based on digital twin as described in claim 4, characterized in that, The generation of the global state transition probability matrix describing the topological transition correlation includes: Construction dimension Global state transition probability matrix ,in, The number of nodes in the structural stiffness topology diagram is represented by the number of nodes in the structural stiffness topology diagram. In the structural stiffness topology diagram, the state transition probability of a node flowing to its directly connected node is set to the value of the corresponding element in the global state transition probability matrix.

6. The method for dynamic visualization of substation structural status based on digital twin as described in claim 5, characterized in that, The method for calculating the state transition probability from a node to its directly connected nodes is as follows: Set the state transition probability of all stresses that are positively deviating from the node to 1, and set the state transition probability of all stresses that are negatively deviating from the node to 0. For all nodes other than the nodes with positive and negative stress deviations, calculate the sum of the axial deformation stiffness of the steel components between all directly connected nodes of each node, and denot it as the total connected stiffness of each node; take the proportion of the axial deformation stiffness of the steel components between each node and any of its directly connected nodes in the total connected stiffness as the state transition probability of each node flowing to any of its directly connected nodes.

7. The method for dynamic visualization of substation structural status based on digital twin as described in claim 6, characterized in that, The process of generating the directed acyclic graph is as follows: Extract the row vectors of the rows where the stress negative deviation nodes are located from the expected path number matrix, and use the column indexes of the elements with values ​​greater than 0 in the row vectors as the corresponding nodes as transit nodes to construct an undirected topology subnet. In an undirected topological subnet, for any two directly interconnected nodes, the values ​​of the corresponding values ​​of the two nodes in the expected number of traversals matrix are compared. The connection direction of the edge between the two nodes is defined as the node with the larger value pointing to the node with the smaller value, so that the undirected topological subnet is transformed into a directed acyclic graph without closed loops.

8. The method for dynamic visualization of substation structural status based on digital twin as described in claim 7, characterized in that, The process of extracting the load transfer spatial connectivity path directly connecting the start and end points using depth-first search includes: In a directed acyclic graph, the search starts at nodes with negative stress deviation and ends at nodes with positive stress deviation. A depth-first search algorithm is executed to eliminate isolated edges and invalid connected edges in the directed acyclic graph that cannot reach the endpoint, thereby generating a connected path for load transfer space.

9. The method for dynamic visualization of substation structural status based on digital twin as described in claim 1, characterized in that, The driving digital twin screen visualizes the spatial connectivity path of load transfer, including: The load transfer spatial connectivity path is mapped to the spatial coordinate system of the substation's 3D model and displayed in a dynamically highlighted manner.

10. A dynamic visualization system for the structural status of a substation based on digital twins, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the substation structure status dynamic visualization method based on digital twin as described in any one of claims 1-9.

Images