Wind farm profile real-time reconstruction method and system

Abstract

The present application relates to the technical field of wind farm monitoring and management, and specifically discloses a wind farm profile virtual-real fusion real-time reconstruction method and system. It includes five stages: first, the task is adaptively distributed through the heterogeneous edge-cloud collaborative computing node cluster; second, the wind farm is divided into semantic paragraphs and marked with value based on the characteristics of terrain, wake and power gradient; third, the semantic fracture index is calculated to judge the profile stability, and the reconstruction instruction is generated when needed; fourth, the differentiated visual reconstruction is performed according to the value label, and the double-path robustness check is carried out in parallel; finally, the system parameters are optimized in a closed loop combined with the interactive feedback and the check results. This method realizes the whole process from data acquisition, semantic understanding, intelligent reconstruction to continuous optimization, and improves the real-time performance, accuracy and adaptive ability of the wind farm visualization analysis.

Description

Technical Field

[0001] This invention relates to the field of wind farm monitoring and management technology, specifically to a method and system for real-time reconstruction of wind farm profiles through virtual-real fusion. Background Technology

[0002] With the acceleration of the global energy transition, wind power is playing an increasingly important role in the electricity mix. However, the operational efficiency of wind farms remains limited by their complex operating environment and inadequate monitoring methods. Traditional wind farm monitoring systems often rely on two-dimensional static charts or simple 3D models with low refresh rates for cross-sectional display, making it difficult to reflect changes in transient physical fields such as turbulence, wake effects from complex terrain, and gusts in real time and intuitively. Studies indicate that wake effects can cause downstream turbine power losses of up to 20%, and traditional static display methods cannot effectively capture its dynamic evolution, leading to delayed operation and maintenance decisions and becoming one of the key bottlenecks restricting the overall power generation efficiency of wind farms.

[0003] At the computational and visualization levels, existing methods often employ a uniform rendering strategy, failing to differentiate the importance of different areas within the wind farm. For example, they allocate the same computational and rendering resources to areas with gentle terrain and weak wake effects as to high-value areas located at canyon entrances with severe wake superposition. An analysis of a typical 100-megawatt wind farm monitoring system shows that over 60% of graphics computing resources are consumed in low-value areas with low information entropy, resulting in insufficient visualization details of high-value information such as power surges and potential mechanical failure precursors, and an increased rate of missed alarms for critical alarm information.

[0004] Current technologies rely heavily on fixed rules or human experience to determine where key monitoring areas should be monitored, lacking adaptive and quantitative identification mechanisms. Although some studies have attempted to introduce machine learning for anomaly detection, these studies mostly focus on single units or single physical quantities, failing to integrate and collaboratively judge multiple semantic dimensions such as terrain, wake, and power gradient, resulting in limited accuracy in identifying high-value areas. Summary of the Invention

[0005] This invention aims to provide a method and system for real-time reconstruction of wind farm profiles through virtual-real fusion, addressing the problems in existing technologies such as static profile display lacking dynamic details, insufficient rendering of key information due to unreasonable allocation of edge and cloud computing resources, and inaccurate identification of high-value areas due to a lack of multi-dimensional quantitative judgment. By dynamically deploying and optimizing heterogeneous edge-cloud collaborative computing tasks in each reconstruction cycle, performing semantic segmentation and three-layer threshold judgment marking on the wind farm profile, and triggering differentiated reconstruction and dual-path verification based on a semantic breakage index, this invention achieves accurate identification of high-value areas in the wind farm profile, stable real-time reconstruction of the virtual-real fusion view, and intelligent optimization of the monitoring and maintenance closed loop.

[0006] The technical solution of this application specifically includes:

[0007] According to one aspect of this application, a method for real-time reconstruction of wind farm profiles through virtual-real fusion is provided, comprising:

[0008] Before each reconstruction cycle begins, a heterogeneous edge-cloud collaborative computing node cluster is deployed to collect real-time status and historical performance. It automatically matches and allocates five types of sub-tasks—data collection, semantic extraction, deviation detection, structural rearrangement, and interactive feedback analysis—to the optimal node and outputs a task allocation scheme.

[0009] With the support of the task allocation scheme, the wind farm is divided into multiple semantic segments along the prevailing wind direction. Each segment encapsulates three types of physical semantic features: terrain, wake, and power gradient. A three-level nested judgment is initiated for the semantic segments: thresholds are set for terrain complexity, wake significance, and power gradient steepness, and comparisons are made. Based on the comparison results, the segments are marked as high-value, medium-value, and low-value regions, and finally, a labeled semantic segment map is output.

[0010] Calculate the semantic breakage index for each region of the semantic segment map. If it is below the threshold, the current profile structure is determined to be stable and a flag indicating that no reconstruction is needed is output. If it exceeds the threshold, a reconstruction instruction package is generated based on the semantic breakage index.

[0011] Based on the reconstruction instruction package, the semantic segment graph is reconstructed, and dual-path verification is run in parallel: one path performs adversarial robustness testing on the profile view through parameter perturbation, and the other path reverses the semantic segment division of the reconstruction instruction to verify whether the current reconstruction result is a necessary solution for the semantic segment. Finally, the verified profile view and verification result report are output.

[0012] If the input is a "no reconstruction required" flag, then the parameters are fine-tuned based on the current system status; if the input is a verified profile view and verification result report, then the interaction trajectory between the operations and maintenance personnel and the profile view is captured, and the system parameters are adjusted in conjunction with the verification result report; after the parameter adjustment is completed, the system status is updated and the system waits for the next reconstruction cycle to be triggered.

[0013] As a further option of the method of the present invention, the steps of deploying a heterogeneous edge-cloud collaborative computing node cluster to collect real-time status and historical performance, and automatically matching and allocating five types of sub-tasks to the optimal node include:

[0014] The central management platform periodically sends status query commands to each computing node in the cluster. After receiving the command, the node collects local CPU utilization, memory utilization, remaining network bandwidth percentage, average network latency between the node and the wind farm data source, and node type identifier, encapsulates them into a status vector, and reports them.

[0015] The central management platform maintains a database of node historical performance, recording and statistically analyzing the historical average execution efficiency and result reliability scores of each node for different types of tasks.

[0016] Based on the reported real-time status vectors and historical performance data, the central management platform calculates the real-time comprehensive performance evaluation value for each node; the real-time comprehensive performance evaluation value is obtained by weighted synthesis of real-time load factor, network condition factor and historical performance factor;

[0017] Based on real-time comprehensive performance evaluation values ​​and preset task-specific preference factors, the system constructs a fitness matrix from task type to computing node; using the fitness matrix as the benefit matrix, it employs an auction algorithm to solve for a task allocation scheme that maximizes the total fitness, and then distributes the scheme to each computing node.

[0018] As a further option of the method of the present invention, the step of marking regions as high-value, medium-value, and low-value based on the comparison results includes:

[0019] High-value judgment thresholds are set for three types of physical semantic features: terrain complexity, wake significance, and power gradient steepness.

[0020] Perform the following steps sequentially for each semantic paragraph:

[0021] First-level judgment: Determine whether the terrain complexity of the semantic paragraph is greater than or equal to the terrain complexity threshold. If yes, record the first flag as 1; otherwise, record it as 0.

[0022] The second layer of judgment: Determine whether the tail flow salience of the semantic paragraph is greater than or equal to the tail flow salience threshold. If yes, record the second flag bit as 1; otherwise, record it as 0.

[0023] The third layer of judgment: Determine whether the power gradient steepness of the semantic paragraph is greater than or equal to the power gradient steepness threshold. If yes, record the third flag bit as 1; otherwise, record it as 0.

[0024] Add the first, second, and third flags of the semantic paragraph to obtain the number of the semantic paragraph's compliance level;

[0025] If the number of qualified levels is 3, the semantic paragraph is marked as a high-value area; if the number of qualified levels is 2, the semantic paragraph is marked as a medium-value area; if the number of qualified levels is less than or equal to 1, the semantic paragraph is marked as a low-value area.

[0026] As a further option of the method of the present invention, the step of dividing the wind farm into multiple semantic segments along the prevailing wind direction, and encapsulating three types of physical semantic features—topography, wake, and power gradient—in each segment includes:

[0027] Acquire digital elevation model data of the wind farm, wind turbine layout coordinates, and the current prevailing wind direction vector;

[0028] Starting from the boundary of the wind farm inlet, the wind farm area is divided into a series of continuous strips perpendicular to the prevailing wind direction vector, with each strip defined as a semantic segment.

[0029] For each semantic segment, data acquisition and feature extraction subtasks are executed in parallel. The subtasks are scheduled to the optimal computing node according to the task allocation scheme to extract and encapsulate the topographic complexity, wake significance and power gradient steepness of the semantic segment.

[0030] As a further option of the method of the present invention, the formula for calculating the semantic breakage index is: ;in, and These are the weighting coefficients; For value label jump penalty items, , The penalty coefficient is... This represents the absolute value of the difference in tag ordinal numbers between the current semantic paragraph and the upstream semantic paragraph. This represents the absolute value of the difference in tag ordinal numbers between the current semantic paragraph and the downstream semantic paragraph. For the physical feature gradient magnitude term, , The feature vector difference between the current semantic paragraph and the upstream semantic paragraph. The feature vector difference between the current semantic paragraph and the downstream semantic paragraph. This represents the L2 norm of a vector.

[0031] As a further option of the method of the present invention, the step of generating a reconstruction instruction package based on the semantic breakage index includes:

[0032] Identify all semantic segments in the semantic segment map whose semantic breakage index exceeds a preset breakage threshold, and form a set of key breakage regions.

[0033] For each semantic segment in the set of key fracture regions, the reconstruction operation type is specified based on the value tag that the segment has been marked in the semantic segment map. The reconstruction operation type includes: if it is a high-value region, the operation type is enhanced rendering; if it is a medium-value region, the operation type is standard rendering; if it is a low-value region, the operation type is compression and simplification.

[0034] For semantic paragraphs whose operation type is specified as enhanced rendering, a spatial stretching factor based on the paragraph semantic breakage index mapping is introduced;

[0035] The set of key fracture regions, the reconstruction operation type and parameters corresponding to each region are encapsulated into a structured reconstruction instruction package.

[0036] As a further option of the method of the present invention, reconstructing the semantic paragraph graph according to the reconstruction instruction package includes:

[0037] The rendering engine receives semantic paragraph graphs and reconstruction instruction packets;

[0038] Traverse each semantic segment in the semantic segment graph and check the value tag of the semantic segment and the specified operation type in the reconstruction instruction package:

[0039] If a semantic paragraph is marked as a high-value region and the refactoring instruction package is specified as enhanced rendering, then spatial stretching is performed on the paragraph, a high-resolution model is invoked for detail enhancement, and dynamic particle tracing driven by computational fluid dynamics simulation data is overlaid.

[0040] If a semantic paragraph is marked as a low-value region and the refactoring instruction package is specified as compressed and simplified, then the paragraph is subjected to fan icon aggregation display and low-poly terrain simplification rendering.

[0041] Perform standard rendering with the system's default precision on other semantic paragraphs.

[0042] As a further option of the method of the present invention, the parallel dual-path verification includes:

[0043] The first path, namely the adversarial robustness test under parameter perturbation, includes: applying random noise to the physical feature vectors in the semantic segment graph to generate a perturbation graph; reconstructing the perturbation graph based on the same reconstruction instruction package to generate a perturbation-reconstructed view; obtaining the difference between the perturbation-reconstructed view and the original reconstruction view, and comparing it with the first robustness threshold to determine whether the test passes.

[0044] The second path is executed, which is the logical necessity verification of the reverse backtracking of the reconstruction instruction. This includes: inferring the value label applied to each semantic segment from the visual style of the generated reconstructed profile view; re-acquiring the theoretical value label of each semantic segment using the original physical feature data and the same three-level nested judgment rules; judging the overall consistency rate between the inferred label and the theoretical label, and comparing it with the second consistency threshold to determine whether the verification passes.

[0045] As a further option of the method of the present invention, the step of capturing the interaction trajectory between the operation and maintenance personnel and the profile view, and adjusting the system parameters in conjunction with the verification result report, includes:

[0046] The system displays the verified profile view through a human-computer interaction interface and captures the sequence of interaction events generated by maintenance personnel;

[0047] Interaction trajectory features are extracted from the sequence of interaction events. These features include dwell time in high-value marked areas, grazing frequency in low-value marked areas, and a set of manually added attention point coordinates.

[0048] Read the conclusions and detailed data from the verification result report;

[0049] If the verification result report concludes as successful, the overlap between the interaction trajectory features and the high-value areas marked by the system is analyzed, and the thresholds for terrain complexity, wake significance, or power gradient steepness involved in the semantic segment value marking are adjusted based on the analysis results.

[0050] If the verification result report concludes as a failure, then by combining the reasons for failure in the report with the characteristics of the interaction trajectory, the weak links in the system are located, and the feature extraction algorithm parameters, task allocation strategies, or judgment thresholds are adjusted.

[0051] Another aspect of this application provides a real-time reconstruction system for virtual-real fusion of wind farm profiles, the system comprising:

[0052] The task coordination and allocation module is used to deploy a heterogeneous edge-cloud collaborative computing node cluster before the start of each reconstruction cycle, collect real-time status data of computing nodes and obtain their historical performance data, automatically match and allocate five types of sub-tasks—data collection, semantic extraction, deviation detection, structural rearrangement, and interactive feedback analysis—to the optimal computing node based on the real-time status data and historical performance data, and output the task allocation scheme.

[0053] The semantic value classification module is used to divide the wind farm area into multiple semantic segments along the prevailing wind direction under the support of the task allocation scheme. It extracts and encapsulates three types of physical semantic features of each semantic segment in parallel: terrain complexity, wake significance, and power gradient steepness. It performs three-level nested judgment on each semantic segment to mark it as a high-value region, medium-value region, or low-value region according to the comparison results of preset thresholds. Finally, it generates and outputs a semantic segment map with value labels.

[0054] The intelligent reconstruction decision module is used to calculate the semantic breakage index of each region in the semantic segment map, compare the semantic breakage index with the preset breakage threshold, and if it is lower than the breakage threshold, the profile structure is determined to be stable and a flag indicating that no reconstruction is needed is output. If it exceeds the breakage threshold, a reconstruction instruction package containing a list of key breakage regions, differentiated reconstruction operation types and related parameters is generated.

[0055] The profile rendering and verification module is used to perform differentiated visual reconstruction of the semantic segment map according to the reconstruction instruction package to generate a preliminary reconstructed profile view, and to perform dual-path robustness verification in parallel: the first path performs adversarial robustness testing by applying perturbation to physical features, and the second path performs logical necessity verification by comparing reverse inference with forward rules, and finally outputs the verified profile view and verification result report.

[0056] The closed-loop optimization and update module is used to receive and determine the input type: if the input is a flag indicating that no reconstruction is needed, then the parameters are fine-tuned based on the current system operating status; if the input is a verified profile view and a verification result report, then the interaction trajectory between the operation and maintenance personnel and the verified profile view is captured, and the system parameters are adaptively adjusted in combination with the verification result report; after the parameter adjustment is completed, the system status is updated, and the system waits for the next reconstruction cycle to be triggered.

[0057] The beneficial effects of this application are as follows:

[0058] This method, by reconstructing the wind farm profile in real time and dynamically calculating the semantic fracture index, can promptly capture the dynamic evolution of the wake effect. This allows operation and maintenance personnel to quickly identify and respond to transient changes in the wind farm structure, potentially mitigating downstream turbine power losses of up to 20% caused by wake effects and improving the overall power generation efficiency of the wind farm.

[0059] This method, through multi-physical feature fusion and a three-layer nested judgment mechanism, enables the system to achieve semantic value classification of wind farm areas and perform differentiated rendering based on value tags. It completely reverses the inefficient situation in traditional monitoring systems where over 60% of graphics computing resources are consumed in low-value areas, precisely focusing core computing power and rendering resources on high-value information, significantly enhancing the visualization details and recognizability of critical alarms.

[0060] This method combines interactive feedback with dual-path verification results for self-optimization, continuously adjusting the value judgment threshold and task allocation strategy. This not only improves the accuracy of identifying high-value areas with complex terrain and significant wakes, but also endows the system with the ability to continuously evolve and adapt, thereby optimizing monitoring performance in the long term and effectively supporting intelligent operation and maintenance decisions for wind farms. Attached Figure Description

[0061] Figure 1 A schematic diagram of the real-time reconstruction method for virtual-real fusion of wind farm profiles;

[0062] Figure 2 S100 flowchart of the real-time reconstruction method for virtual-real fusion of wind farm profiles;

[0063] Figure 3 S200 flowchart of the real-time reconstruction method for virtual-real fusion of wind farm profiles;

[0064] Figure 4 S300 flowchart for real-time reconstruction method of virtual-real fusion of wind farm profile;

[0065] Figure 5 S400 flowchart for real-time reconstruction method of virtual-real fusion of wind farm profile;

[0066] Figure 6 The flowchart of the S500 method for real-time reconstruction of wind farm profiles through virtual-real fusion. Detailed Implementation

[0067] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0068] The core theoretical foundation of this invention is built upon the theories of real-time processing and scheduling of high-dimensional data streams, multimodal semantic feature fusion and decision-making, and closed-loop verification and optimization driven by digital twins. By constructing a task adaptive allocation model in a heterogeneous computing environment, designing a regional value grading mechanism based on multi-semantic features and a three-layer nested criterion, and introducing a semantically fracture-driven intelligent reconstruction triggering and dual-path robust verification strategy, a complete link is ultimately formed from task collaboration, semantic understanding, intelligent reconstruction to interactive optimization.

[0069] The specific embodiments of the present invention will be described in detail below.

[0070] Example 1:

[0071] Please see Figure 1 , Figure 1 This diagram illustrates the overall flowchart of a real-time reconstruction method for virtual-real fusion of wind farm profiles provided by an embodiment of the present invention. The method includes:

[0072] S100: Deployment and adaptive task allocation phase of heterogeneous edge-cloud collaborative computing node clusters;

[0073] S200: Wind farm semantic segmentation, physical feature encapsulation and regional value marking stage;

[0074] S300: The stage of judging the stability of the profile structure and generating reconstruction instructions based on the semantic fracture index;

[0075] S400: Differentiated profile reconstruction execution and dual-path robustness verification phase;

[0076] S500: Interactive feedback-driven system parameter fine-tuning and state update phase.

[0077] The following provides a detailed explanation of each stage.

[0078] In a real-time reconstruction method for virtual-real fusion of wind farm profiles, S100 dynamically deploys and evaluates heterogeneous computing clusters before the start of each reconstruction cycle, thereby achieving automatic matching and allocation of five key sub-tasks to the optimal computing nodes, laying an efficient and collaborative computing foundation for subsequent processing.

[0079] Please refer to Figure 2 , Figure 2 A flowchart of stage S100 in an exemplary embodiment of this application is shown.

[0080] S110: In this embodiment, the system maintains a heterogeneous computing resource pool that includes edge computing nodes and cloud virtual machines. Edge nodes are deployed inside wind farm booster stations or key wind turbine towers, responsible for low-latency data acquisition and preliminary processing; cloud nodes have powerful computing capabilities, responsible for complex semantic analysis and global optimization.

[0081] During the implementation of this step, the dynamic perception and state modeling of the heterogeneous edge-cloud collaborative computing node cluster includes:

[0082] S111: Periodically, the central management platform sends data to each compute node in the cluster. Send a status query command. Node Upon receiving the instruction, collect local multidimensional state data and encapsulate it into a state vector. Report it to higher authorities.

[0083] In one possible implementation of this embodiment, the state vector Includes the following field: Node identifier Current CPU utilization Current memory utilization Current network bandwidth remaining percentage Average network latency between nodes and wind farm data sources and node type identifier ,in, Represents edge nodes. This represents a cloud node.

[0084] S112: The central management platform maintains a database of node historical performance. Whenever a computational task occurs on a node... Once the task is completed, the platform records the task type, input data volume, execution time, and final result accuracy. By statistically analyzing historical execution records of similar tasks on the same node, the node's... Based on task type Historical average execution efficiency and result reliability score .

[0085] S113: To quantify the real-time overall performance of nodes and provide a basis for task allocation, the system calculates the performance of each node. Real-time comprehensive performance evaluation value . The calculation combines real-time status and historical performance, and its formula is as follows: ;in, For adjustable weighting coefficients, satisfying This is used to balance the importance of real-time resources, network conditions, and historical performance. It is based on the task type The weight, . in the formula The item will translate high load into low efficiency contribution. This allows low-latency nodes to receive higher scores.

[0086] S120: In this embodiment, the five types of sub-tasks are defined as: data acquisition Semantic extraction Deviation detection Structural rearrangement and interactive feedback analysis .

[0087] During this step, the generation of optimal node matching and allocation schemes for the five types of subtasks includes:

[0088] S121: System build task type To compute node fitness matrix Elements in the matrix Indicates task type At the node The expected fit of the execution. The calculation is based not only on the overall performance of the nodes It also introduced task-specific preference factors. . The calculation formula is: ,in This is the global balance coefficient.

[0089] S122: The system constructs a bipartite graph matching problem by matching the five types of subtask instances that need to be executed within the current reconstruction cycle with the currently available set of computing nodes. The fitness matrix is ​​used as the basis for this matching. As the benefit matrix, an auction algorithm is used to solve for the task allocation scheme that maximizes the overall fitness. ,in Indicates the task instance Assigned to node implement.

[0090] S123: The central management platform will finalize the task allocation plan. The allocation plan is distributed to each computing node. Each node loads the corresponding task processing program and model according to the allocation plan, preparing to receive data and execute tasks. The platform synchronously outputs this allocation plan as the control basis for the subsequent data processing flow in the S200 phase.

[0091] In a real-time reconstruction method for virtual-real fusion of wind farm profiles, S200 performs semantic segmentation of the wind farm based on the task allocation scheme output by S100, extracts key physical features, and assigns value tags to each segment through a three-layer nested judgment rule to generate a semantic segment map with value tags.

[0092] Please refer to Figure 3 , Figure 3 A flowchart of stage S200 in an exemplary embodiment of this application is shown.

[0093] S210: In this embodiment, the system acquires the digital elevation model data of the wind farm, the wind turbine layout coordinates, and the current prevailing wind direction vector. Starting from the boundary of the wind farm inlet, along the prevailing wind direction vector... The direction divides the wind farm area into a series of continuous, perpendicular sections. The stripes. Each stripe is defined as a semantic segment. , .

[0094] S220: For each semantic paragraph The task of extracting three types of physical semantic features is executed in parallel, and the extraction task is assigned to the optimal node for execution according to the S100 scheme.

[0095] During this step, the physical characteristics of the semantic paragraph include:

[0096] S221: Terrain Complexity Quantification The degree to which inland ground undulations interfere with wind flow.

[0097] S222: Wake Significance Quantification The severity of the wake effect of the internal fan.

[0098] S223: Power gradient steepness Quantification The degree of drastic change in the output power of the internal fan.

[0099] S230: In this embodiment, the terrain is complex. Wake salience and power gradient steepness Set high-value judgment thresholds respectively , , .

[0100] In this step, the three-level nested judgment and value marking process includes:

[0101] S231: The first level of judgment is based on whether the terrain complexity meets the standard. Judgment conditions: If true, record the flag bit. ,otherwise .

[0102] S232: The second level of judgment is based on whether the wake significance meets the standard. Judgment conditions: If true, record the flag bit. ,otherwise .

[0103] S233: The third layer judgment is based on whether the power gradient steepness meets the standard. Judgment conditions: If true, record the flag bit. ,otherwise .

[0104] S234: Calculating semantic paragraphs The number of qualified layers .

[0105] like Then Marked as a high-value area.

[0106] like Then Marked as a medium-value area.

[0107] like Then Marked as a low-value area.

[0108] S240: The system completes the semantic paragraphs with all tags. and their corresponding value tags and original physical characteristic values Including spatial location information, it integrates to generate structured data objects, namely labeled semantic segment maps. And output it to the subsequent stages.

[0109] In a real-time reconstruction method for virtual-real fusion of wind farm profiles, S300 performs structural stability assessment on semantic segment maps, determines whether profile reconstruction is needed by calculating semantic fracture index, and generates precise reconstruction instruction packages when necessary.

[0110] Please refer to Figure 4, Figure 4 A flowchart of stage S300 in an exemplary embodiment of this application is shown.

[0111] S310: In this embodiment, the system reads the semantic segment graph. For semantic paragraph maps Each semantic paragraph Extracting value tags and physical feature vectors At the same time, define The upstream and downstream adjacent sections. Along the prevailing wind direction, Its upstream directly adjacent segment, It is the directly adjacent paragraph downstream of it.

[0112] S320: Semantic Breakage Index Used for quantification The degree to which the semantic continuity of the wind field profile is interrupted or abruptly changed.

[0113] During this step, the semantic breakage index calculation includes:

[0114] S321: Define the ordinal mapping of value tags, where low value is mapped to 1, medium value to 2, and high value to 3. Calculate... Upstream absolute value of the difference in label ordinal numbers Similarly, computation and downstream The difference .

[0115] Value tag jump penalty item Defined as: ,in This is the penalty coefficient.

[0116] S322: Calculation Upstream eigenvector difference .

[0117] calculate Downstream eigenvector difference .

[0118] Physical feature gradient magnitude term Defined as the average of the magnitudes of two difference vectors: ,in This represents the L2 norm of a vector.

[0119] S323: Semantic Paragraph semantic break index The above two items are combined by weighting: ,in and This is a weighting coefficient used to balance the effects of label mutations and feature changes.

[0120] S330: System preset global semantic break threshold .

[0121] During this step, global stability checks and branch handling include:

[0122] S331: Computational Graph Find the semantic breakage index of all semantic paragraphs and identify the maximum value. .

[0123] S332: Comparison and .like If the current wind farm profile structure is stable overall, then no large-scale reconstruction is required. The system generates a "no reconstruction required" flag. And then jump directly to the S500 stage.

[0124] like If the current profile structure is unstable, it is determined that there are semantic breakpoints that need to be reconstructed. The system then enters S340.

[0125] S340: When a refactoring is determined to be necessary, the system generates a detailed refactoring instruction package. .

[0126] During this step, the generation of the refactoring instruction package includes:

[0127] S341: Find all that satisfy the condition semantic paragraphs Semantic paragraph The key fracture areas are marked as a set. .

[0128] S342: For each critical fracture region The type of reconstruction operation is specified based on the value tag obtained in S200.

[0129] like For high-value areas, specify the operation type as enhanced rendering.

[0130] like If the value region is specified, then the operation type is set to standard rendering.

[0131] like If the region is of low value, then the operation type is specified as compression and simplification.

[0132] For non-critical fracture areas, the default rendering operation is also specified based on their value tags.

[0133] S343: For enhanced rendering areas, a spatial stretching factor is introduced. . based on The magnitude is obtained by linear or nonlinear mapping. Other parameters, such as the intensity of detail enhancement and the ratio of compression simplification, are also introduced parametrically based on the value tag and fracture index.

[0134] S344: List of key fracture areas The reconstruction operation type for each region, the reconstruction parameters introduced, and the entire graph. The metadata is collectively encapsulated into a structured refactoring instruction package. Output to the S400 stage.

[0135] In a real-time reconstruction method for virtual-real fusion of wind farm profiles, the S400 performs differentiated visual reconstruction of semantic segment maps based on reconstruction instruction packages, and performs dual-path verification in parallel to ensure the robustness and logical necessity of the reconstruction results.

[0136] Please refer to Figure 5 , Figure 5 A flowchart of stage S400 in an exemplary embodiment of this application is shown.

[0137] S410: In this embodiment, the rendering engine receives the semantic segment graph. and refactoring instruction package The engine traverses each semantic segment in the graph. .

[0138] During the implementation of this step, the value-label-based differentiated profile reconstruction includes:

[0139] S411: For For high value and being Paragraphs designated for enhanced rendering include:

[0140] Perform spatial stretching: In the visualization view, expand the display width of the paragraph along a direction perpendicular to the prevailing wind direction. times.

[0141] Enhanced execution details: High-resolution terrain textures, refined wake vortex visualization models, and gradient coloring algorithms for power curves are invoked to improve visual information density.

[0142] Perform dynamic tracing: Overlay a set of time-rolling particle streamlines onto the region, with particle velocity vectors driven by instantaneous flow field data from computational fluid dynamics simulations.

[0143] S412: For For low value and being Paragraphs designated for compression and simplification include:

[0144] Perform visual compression: Aggregate all fan icons in a paragraph into a group icon, and only display the average power and number of fans in the group.

[0145] Perform terrain simplification: Use a low-poly model instead of a high-precision digital elevation model to render the area terrain.

[0146] S413: For For medium value or not Specific paragraphs are rendered using standard rendering, which means they are drawn using the system's default rendering precision and visual style.

[0147] S414: After completing the differential rendering of all semantic paragraphs, the engine generates a complete, preliminary reconstructed profile view. .

[0148] S420: To ensure the reliability of the reconstruction results, the system starts two verification paths in parallel.

[0149] During this step, two verification paths are included:

[0150] S421: Path 1: Adversarial robustness test under parameter perturbation.

[0151] S421a: For the original semantic segment map Physical feature vectors in Apply small random noise Generate perturbation map .

[0152] S421b: Uses the same refactoring instruction package However, based on Execute the differential rendering logic in S410 to generate a perturbed and reconstructed view. .

[0153] S421c: Calculation and Difference at the pixel level or feature level In one possible implementation, the degree of difference It is calculated using a structural similarity index or feature matching error.

[0154] S421d: Set robustness threshold .like Then it is considered that the view is reconstructed. The adversarial robustness test has been passed; otherwise, path one verification would fail.

[0155] S422: Path 2: Verification of the logical necessity of reconstructing instructions by backtracking backward.

[0156] S422a: Through analysis The visual styles of each region are used to infer the semantic paragraphs. The value tags applied in this restructuring .

[0157] S422b: Raw physical characteristic data saved using the S200 stage With the same three-level nested judgment rule, recalculate each Theoretical value label .

[0158] S422c: Value Labels Based on Reverse Inference Value labels of positive reasoning Calculate the label consistency rate: ;in, This is an indicator function; it returns 1 if the two labels are the same, and 0 otherwise. This represents the total number of paragraphs.

[0159] S422d: Set consistency threshold ,like If the result is correct, then the current reconstruction result is considered a necessary solution for the semantic features; otherwise, the path two verification fails.

[0160] S430: The system waits for the two verification paths to complete before integrating and outputting the verification results.

[0161] During this step, the integration and output of verification results include:

[0162] S431: The report should include: the test results and differences for Path 1. Specific numerical values; verification results and consistency rate of Path 2. Specific numerical values; comprehensive verification conclusions.

[0163] S432: The system will verify the cross-sectional view. Verification Result Report The common output serves as the input for the S500 stage.

[0164] In a real-time reconstruction method for virtual-real fusion of wind farm profiles, the S500 performs fine-tuning of system parameters according to different inputs, updates the system status based on interactive feedback from maintenance personnel, completes the closed loop of the reconstruction cycle, and waits for the next cycle to be triggered.

[0165] Please refer to Figure 6 , Figure 6A flowchart of stage S500 in an exemplary embodiment of this application is shown.

[0166] S510: There are two possible inputs for the S500 stage: one is the no-reconfiguration flag from S330. Second, verified cross-sectional views from S430. Verification Result Report .

[0167] During this step, input type judgment and branch processing include:

[0168] S511: If the input is If the input is..., then jump to S520. If so, it will jump to S530.

[0169] S520: When the cross-sectional structure is stable and does not require reconstruction, the system enters the normal optimization mode and fine-tunes the parameters based on the stable state.

[0170] During this step, parameter fine-tuning based on the steady state includes:

[0171] S521: The system collects routine performance metrics for the current period, including: average node efficiency of task allocation. Average time spent on semantic feature extraction Frame rate of view rendering .

[0172] S522: Compare the above performance indicators with the preset target range.

[0173] like If the value is lower than the target value, the weighting coefficients in S100 are fine-tuned. .

[0174] like If the target value is exceeded, the balance parameters between the computational accuracy and speed of the feature extraction algorithm in S200 are fine-tuned.

[0175] like If the value is lower than the target value, the global parameters of the rendering engine at the detail level in S400 are fine-tuned.

[0176] The fine-tuning process employs lightweight optimization algorithms based on gradient descent or heuristic rules.

[0177] S523: Record the parameters modified in this fine-tuning, the reasons for the modification, and the expected improvement effect. Then jump to S540.

[0178] S530: After the profile is reconstructed and verified, the system enters an interactive enhancement optimization mode to adjust system parameters through interactive feedback and verification reports.

[0179] During this step, the system parameter adjustments for interactive feedback and verification reports include:

[0180] S531: The system provides a human-computer interaction interface. Operations and maintenance personnel may perform interactive operations such as zooming in and out of specific areas, clicking to query details, dragging and dropping views, and marking points of interest. The system captures and records all interaction event sequences within a time window, extracting interaction trajectory features, such as the duration of time spent in high-value areas. Frequency of rapid traversal in low-value areas The set of manually marked points of interest wait.

[0181] S532: Read the verification result report If the report concludes that the verification was successful, it indicates that the reconstruction result is reliable. Analysis of interaction patterns: If the areas frequently interacted by operations and maintenance personnel highly overlap with the high-value areas marked by the system, the effectiveness of the S200 value marking rules is verified; if operations and maintenance personnel show abnormally high attention to a certain medium-value area, it may indicate that the value judgment threshold for that area needs to be adjusted.

[0182] If the report concludes that the verification failed, it indicates that there is a problem with the reconstruction results. By combining the reasons for the failure in the report and analyzing the interaction trajectory, we can jointly pinpoint the weak points in the system.

[0183] S533: Based on the in-depth analysis results of S531 and S532, the system parameters are adjusted in a more targeted manner.

[0184] In one possible implementation, the adjustment includes:

[0185] If the interaction trajectory shows high attention to a certain type of feature but the system does not adequately mark it as high value, then adjust S230. , , Equal threshold.

[0186] If path 1 verification fails frequently, adjust the random noise in S421. The range of magnitude, or adjustment of the task allocation strategy in S100.

[0187] If path 2 verification fails, check the consistency of the feature extraction algorithm in S200, or calibrate the threshold of the three-layer judgment.

[0188] S534: Record the process of this interactive analysis, the conclusions of the verification report on which it was based, the specific parameters adjusted, and the decision-making logic for the adjustment. Then jump to S540.

[0189] S540: Regardless of whether the path is S520 or S530, this step should be performed after the parameter adjustment is completed.

[0190] This step includes the following steps:

[0191] S541: Archive all new data generated during this round of reconstruction to the historical database.

[0192] Update the versions of all relevant models and configuration files in the system according to the latest parameter settings.

[0193] S542: The system enters sleep or low-power monitoring mode, continuously monitoring the wind farm data stream. The next reconstruction cycle can be triggered by a fixed time interval or dynamically by an external event. When the triggering conditions are met, the system restarts from S100, initiating a new virtual-real fusion real-time reconstruction cycle.

[0194] In summary, this method achieves adaptive and efficient allocation of wind farm data processing tasks by constructing a heterogeneous edge-cloud collaborative computing framework; it completes semantic value classification of wind farm profiles by extracting multiple physical features such as terrain, wake, and power gradient and designing three-layer nested judgment rules; it realizes intelligent judgment of profile structural stability and accurate generation of reconstruction instructions by defining and calculating a semantic fracture index; it ensures the quality and reliability of the reconstructed view by performing value-label-based differentiated rendering and parallel dual-path robustness verification; and finally, it achieves closed-loop self-optimization of system parameters by integrating stable state monitoring, interactive behavior analysis, and verification result feedback. This method significantly improves the real-time performance, accuracy, interpretability, and adaptability of wind farm profile visualization, providing powerful virtual-real fusion decision support for intelligent operation and maintenance of wind farms.

[0195] Example 2:

[0196] This invention was fully deployed and verified in a large coastal wind farm, and compared with traditional static monitoring methods and basic dynamic rendering methods. The implementation environment and key parameter configurations are shown in the table below:

[0197] Table 1: System Deployment Environment and Configuration Parameters

[0198]

[0199] During the three-month testing period, the system underwent various typical wind conditions and fault scenarios. To quantitatively evaluate the effectiveness of this invention, multiple sets of comparative experiments were designed, and the comparison results of key performance indicators are as follows:

[0200] Table 2: Comparison of Key Performance Indicators Experiment Results

[0201]

[0202] Table 3: Performance Analysis of Typical Test Scenarios

[0203]

[0204] In particular, the system of this invention played a crucial role in a real-world gearbox temperature anomaly warning event. The temperature of a certain wind turbine gearbox slowly increased, causing a slight change in its output power characteristics, thus altering its power gradient with that of adjacent downstream wind turbines. The system was slightly enlarged. Although this did not trigger a separate system refactoring, the operations personnel zoomed in on the area where the wind turbine was located to view details during the S500 interaction analysis. The system captured this interaction and subsequently slightly reduced the value of that area through the S533 logic in a later cycle. Judgment threshold The system demonstrated its sensitivity to the problem. A few days later, when the wind turbine's power fluctuations intensified, the system marked the affected section as a medium-value area and provided visual cues two refactoring cycles in advance. This ultimately helped maintenance personnel identify early potential faults and prevent unplanned downtime.

[0205] Example 3:

[0206] A real-time reconstruction system for virtual-real fusion of wind farm profiles includes:

[0207] The task coordination and allocation module is used to deploy a heterogeneous edge-cloud collaborative computing node cluster before the start of each reconstruction cycle, collect real-time status data of computing nodes and obtain their historical performance data, automatically match and allocate five types of sub-tasks—data collection, semantic extraction, deviation detection, structural rearrangement, and interactive feedback analysis—to the optimal computing node based on the real-time status data and historical performance data, and output the task allocation scheme.

[0208] The semantic value classification module is used to divide the wind farm area into multiple semantic segments along the prevailing wind direction under the support of the task allocation scheme. It extracts and encapsulates three types of physical semantic features of each semantic segment in parallel: terrain complexity, wake significance, and power gradient steepness. It performs three-level nested judgment on each semantic segment to mark it as a high-value region, medium-value region, or low-value region according to the comparison results of preset thresholds. Finally, it generates and outputs a semantic segment map with value labels.

[0209] The intelligent reconstruction decision module is used to calculate the semantic breakage index of each region in the semantic segment map, compare the semantic breakage index with the preset breakage threshold, and if it is lower than the breakage threshold, the profile structure is determined to be stable and a flag indicating that no reconstruction is needed is output. If it exceeds the breakage threshold, a reconstruction instruction package containing a list of key breakage regions, differentiated reconstruction operation types and related parameters is generated.

[0210] The profile rendering and verification module is used to perform differentiated visual reconstruction of the semantic segment map according to the reconstruction instruction package to generate a preliminary reconstructed profile view, and to perform dual-path robustness verification in parallel: the first path performs adversarial robustness testing by applying perturbation to physical features, and the second path performs logical necessity verification by comparing reverse inference with forward rules, and finally outputs the verified profile view and verification result report.

[0211] The closed-loop optimization and update module is used to receive and determine the input type: if the input is a flag indicating that no reconstruction is needed, then the parameters are fine-tuned based on the current system operating status; if the input is a verified profile view and a verification result report, then the interaction trajectory between the operation and maintenance personnel and the verified profile view is captured, and the system parameters are adaptively adjusted in combination with the verification result report; after the parameter adjustment is completed, the system status is updated, and the system waits for the next reconstruction cycle to be triggered.

[0212] Those skilled in the art will understand that the embodiments of this application are provided as methods, systems, or computer program products. Therefore, this application takes the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application takes the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer program code. The solutions in the embodiments of this application are implemented using various computer languages, exemplified by the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0213] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams.

[0214] These computer program instructions are also stored in a computer read-memory that can direct a computer or other programmed data processing device to operate in a particular manner, such that the instructions stored in the computer read-memory produce an article of manufacture including instruction means that implement the functions specified in the flowchart or multiple flowcharts and / or block diagram blocks or multiple block diagrams.

[0215] These computer program instructions are also loaded onto a computer or other programming data processing device to cause a series of operational steps to be performed on the computer or other programming device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programming device, provide steps for implementing the functions specified in the flowchart flow or multiple flows and / or the block diagram blocks or multiple blocks.

[0216] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0217] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for real-time reconstruction of wind farm profiles using virtual-real fusion, characterized in that, include: Before each reconstruction cycle begins, a heterogeneous edge-cloud collaborative computing node cluster is deployed to collect real-time status and historical performance. It automatically matches and allocates five types of sub-tasks—data collection, semantic extraction, deviation detection, structural rearrangement, and interactive feedback analysis—to the optimal node and outputs a task allocation scheme. With the support of the task allocation scheme, the wind farm is divided into multiple semantic segments along the prevailing wind direction. Each segment encapsulates three types of physical semantic features: terrain, wake, and power gradient. A three-level nested judgment is initiated for the semantic segments: thresholds are set for terrain complexity, wake significance, and power gradient steepness, and comparisons are made. Based on the comparison results, the segments are marked as high-value, medium-value, and low-value regions, and finally, a labeled semantic segment map is output. Calculate the semantic breakage index for each region of the semantic segment map. If it is below the threshold, the current profile structure is determined to be stable and a flag indicating that no reconstruction is needed is output. If it exceeds the threshold, a reconstruction instruction package is generated based on the semantic breakage index. Based on the reconstruction instruction package, the semantic segment graph is reconstructed, and dual-path verification is run in parallel: one path performs adversarial robustness testing on the profile view through parameter perturbation, and the other path reverses the semantic segment division of the reconstruction instruction to verify whether the current reconstruction result is a necessary solution for the semantic segment. Finally, the verified profile view and verification result report are output. If the input is a flag indicating no reconstruction is needed, then the parameters will be fine-tuned based on the current system state. If the input is a verified profile view and a verification result report, the interaction trajectory between the operations and maintenance personnel and the profile view is captured, and the system parameters are adjusted in conjunction with the verification result report; after the parameter adjustment is completed, the system status is updated and the system waits for the next reconstruction cycle to be triggered.

2. The real-time reconstruction method for wind farm profile fusion according to claim 1, characterized in that, The steps for deploying a heterogeneous edge-cloud collaborative computing node cluster to collect real-time status and historical performance data, and automatically matching and assigning five types of subtasks to the optimal node include: The central management platform periodically sends status query commands to each computing node in the cluster. After receiving the command, the node collects local CPU utilization, memory utilization, remaining network bandwidth percentage, average network latency between the node and the wind farm data source, and node type identifier, encapsulates them into a status vector, and reports them. The central management platform maintains a database of node historical performance, recording and statistically analyzing the historical average execution efficiency and result reliability scores of each node for different types of tasks. Based on the reported real-time status vectors and historical performance data, the central management platform calculates the real-time comprehensive performance evaluation value for each node; the real-time comprehensive performance evaluation value is obtained by weighted synthesis of real-time load factor, network condition factor and historical performance factor; Based on real-time comprehensive performance evaluation values ​​and preset task-specific preference factors, the system constructs a fitness matrix from task type to computing node; using the fitness matrix as the benefit matrix, it employs an auction algorithm to solve for a task allocation scheme that maximizes the total fitness, and then distributes the scheme to each computing node.

3. The real-time reconstruction method for wind farm profile fusion according to claim 1, characterized in that, The step of marking regions as high-value, medium-value, and low-value based on the comparison results includes: High-value judgment thresholds are set for three types of physical semantic features: terrain complexity, wake significance, and power gradient steepness. Perform the following steps sequentially for each semantic paragraph: First-level judgment: Determine whether the terrain complexity of the semantic paragraph is greater than or equal to the terrain complexity threshold. If yes, record the first flag as 1; otherwise, record it as 0. The second layer of judgment: Determine whether the tail flow salience of the semantic paragraph is greater than or equal to the tail flow salience threshold. If yes, record the second flag bit as 1; otherwise, record it as 0. The third layer of judgment: Determine whether the power gradient steepness of the semantic paragraph is greater than or equal to the power gradient steepness threshold. If yes, record the third flag bit as 1; otherwise, record it as 0. Add the first, second, and third flags of the semantic paragraph to obtain the number of the semantic paragraph's compliance level; If the number of qualified levels is 3, the semantic paragraph is marked as a high-value area; if the number of qualified levels is 2, the semantic paragraph is marked as a medium-value area; if the number of qualified levels is less than or equal to 1, the semantic paragraph is marked as a low-value area.

4. The real-time reconstruction method for wind farm profile fusion according to claim 3, characterized in that, The steps involved in dividing a wind farm into multiple semantic segments along the prevailing wind direction, and encapsulating three types of physical semantic features—topography, wake, and power gradient—in each segment include: Acquire digital elevation model data of the wind farm, wind turbine layout coordinates, and the current prevailing wind direction vector; Starting from the boundary of the wind farm inlet, the wind farm area is divided into a series of continuous strips perpendicular to the prevailing wind direction vector, with each strip defined as a semantic segment. For each semantic segment, data acquisition and feature extraction subtasks are executed in parallel. The subtasks are scheduled to the optimal computing node according to the task allocation scheme to extract and encapsulate the topographic complexity, wake significance and power gradient steepness of the semantic segment.

5. The real-time reconstruction method for virtual-real fusion of wind farm profiles according to claim 1, characterized in that, The formula for calculating the semantic breakage index is as follows: ;in, and These are the weighting coefficients; For value label jump penalty items, , The penalty coefficient is... This represents the absolute value of the difference in tag ordinal numbers between the current semantic paragraph and the upstream semantic paragraph. This represents the absolute value of the difference in tag ordinal numbers between the current semantic paragraph and the downstream semantic paragraph. For the physical feature gradient magnitude term, , The feature vector difference between the current semantic paragraph and the upstream semantic paragraph. The feature vector difference between the current semantic paragraph and the downstream semantic paragraph. This represents the L2 norm of a vector.

6. The real-time reconstruction method for virtual-real fusion of wind farm profiles according to claim 5, characterized in that, The steps for generating a reconstruction instruction package based on the semantic breakage index include: Identify all semantic segments in the semantic segment map whose semantic breakage index exceeds a preset breakage threshold, and form a set of key breakage regions. For each semantic segment in the set of key fracture regions, the reconstruction operation type is specified based on the value tag that the segment has been marked in the semantic segment map. The reconstruction operation type includes: if it is a high-value region, the operation type is enhanced rendering; if it is a medium-value region, the operation type is standard rendering; if it is a low-value region, the operation type is compression and simplification. For semantic paragraphs whose operation type is specified as enhanced rendering, a spatial stretching factor based on the paragraph semantic breakage index mapping is introduced. The set of key fracture regions, the reconstruction operation type and parameters corresponding to each region are encapsulated into a structured reconstruction instruction package.

7. The real-time reconstruction method for wind farm profile fusion according to claim 1, characterized in that, Reconstructing the semantic paragraph graph according to the reconstruction instruction package includes: The rendering engine receives semantic paragraph graphs and reconstruction instruction packets; Traverse each semantic segment in the semantic segment graph and check the value tag of the semantic segment and the specified operation type in the reconstruction instruction package: If a semantic paragraph is marked as a high-value region and the refactoring instruction package is specified as enhanced rendering, then spatial stretching is performed on the paragraph, a high-resolution model is invoked for detail enhancement, and dynamic particle tracing driven by computational fluid dynamics simulation data is overlaid. If a semantic paragraph is marked as a low-value region and the refactoring instruction package is specified as compressed and simplified, then the paragraph is subjected to fan icon aggregation display and low-poly terrain simplification rendering. Perform standard rendering with the system's default precision on other semantic paragraphs.

8. The method for real-time reconstruction of wind farm profiles by virtual-real fusion according to claim 1, characterized in that, The parallel dual-path verification includes: The first path, namely the adversarial robustness test under parameter perturbation, includes: applying random noise to the physical feature vectors in the semantic segment graph to generate a perturbation graph; reconstructing the perturbation graph based on the same reconstruction instruction package to generate a perturbation-reconstructed view; obtaining the difference between the perturbation-reconstructed view and the original reconstruction view, and comparing it with the first robustness threshold to determine whether the test passes. The second path is executed, which is the logical necessity verification of the reverse backtracking of the reconstruction instruction. This includes: inferring the value label applied to each semantic segment from the visual style of the generated reconstructed profile view; re-acquiring the theoretical value label of each semantic segment using the original physical feature data and the same three-level nested judgment rules; judging the overall consistency rate between the inferred label and the theoretical label, and comparing it with the second consistency threshold to determine whether the verification passes.

9. The real-time reconstruction method for wind farm profile fusion according to claim 1, characterized in that, The steps for capturing the interaction trajectory between maintenance personnel and the profile view, and adjusting system parameters based on the verification result report, include: The system displays the verified profile view through a human-computer interaction interface and captures the sequence of interaction events generated by maintenance personnel; Interaction trajectory features are extracted from the sequence of interaction events. These features include dwell time in high-value marked areas, grazing frequency in low-value marked areas, and a set of manually added attention point coordinates. Read the conclusions and detailed data from the verification result report; If the verification result report concludes as successful, the overlap between the interaction trajectory features and the high-value areas marked by the system is analyzed, and the thresholds for terrain complexity, wake significance, or power gradient steepness involved in the semantic segment value marking are adjusted based on the analysis results. If the verification result report concludes as a failure, then by combining the reasons for failure in the report with the characteristics of the interaction trajectory, the weak links in the system are located, and the feature extraction algorithm parameters, task allocation strategies, or judgment thresholds are adjusted.

10. A real-time reconstruction system for virtual-real fusion of wind farm profiles according to any one of claims 1-9, characterized in that, The system includes: The task coordination and allocation module is used to deploy a heterogeneous edge-cloud collaborative computing node cluster before the start of each reconstruction cycle, collect real-time status data of computing nodes and obtain their historical performance data, automatically match and allocate five types of sub-tasks—data collection, semantic extraction, deviation detection, structural rearrangement, and interactive feedback analysis—to the optimal computing node based on the real-time status data and historical performance data, and output the task allocation scheme. The semantic value classification module is used to divide the wind farm area into multiple semantic segments along the prevailing wind direction under the support of the task allocation scheme. It extracts and encapsulates three types of physical semantic features of each semantic segment in parallel: terrain complexity, wake significance, and power gradient steepness. It performs three-level nested judgment on each semantic segment to mark it as a high-value region, medium-value region, or low-value region according to the comparison results of preset thresholds. Finally, it generates and outputs a semantic segment map with value labels. The intelligent reconstruction decision module is used to calculate the semantic breakage index of each region in the semantic segment map, compare the semantic breakage index with the preset breakage threshold, and if it is lower than the breakage threshold, the profile structure is determined to be stable and a flag indicating that no reconstruction is needed is output. If it exceeds the breakage threshold, a reconstruction instruction package containing a list of key breakage regions, differentiated reconstruction operation types and related parameters is generated. The profile rendering and verification module is used to perform differentiated visual reconstruction of the semantic segment map according to the reconstruction instruction package to generate a preliminary reconstructed profile view. It performs dual-path robustness verification in parallel: the first path performs adversarial robustness testing by applying perturbation to physical features, and the second path performs logical necessity verification by comparing reverse inference with forward rules. Finally, it outputs the verified profile view and verification result report. The closed-loop optimization and update module is used to receive and determine the input type: if the input is a flag indicating that no reconstruction is needed, then the parameters are fine-tuned based on the current system operating status; if the input is a verified profile view and a verification result report, then the interaction trajectory between the operation and maintenance personnel and the verified profile view is captured, and the system parameters are adaptively adjusted in combination with the verification result report; after the parameter adjustment is completed, the system status is updated, and the system waits for the next reconstruction cycle to be triggered.

Images