Rail transit engineering digital twin management and control method based on unmanned aerial vehicle honeycomb inspection

Abstract

The invention relates to a rail transit engineering digital twin management and control method based on unmanned aerial vehicle honeycomb inspection, and solves the problem that a twin model is disjointed from the real-time state of an engineering field, and the method comprises the steps: firstly obtaining multi-source data containing BIM, GIS and Internet of Things data and an unmanned aerial vehicle initial image; carrying out BIM lightweight processing, carrying out registration with a GIS (Geographic Information System) and establishing an engineering entity-environment-event dynamic association network to form a digital twin reference model; based on the real-time state data of the model and external factors, a multi-dimensional evaluation algorithm is used to generate an inspection task priority; an adaptation rule is established in combination with an unmanned aerial vehicle honeycomb performance threshold, and unmanned aerial vehicle obstacle avoidance inspection is scheduled; the inspection data is preprocessed and then fused to the twin model, and the twin model is driven to dynamically iterate so as to realize real-time synchronization with the site. The method has the following effects: real-time and precise rail transit engineering management and control are realized through digital twin virtual-real mapping and unmanned aerial vehicle honeycomb intelligent inspection closed loop, and the core management and control efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent management and control of rail transit engineering construction and operation, and in particular to a digital twin management and control method for rail transit engineering based on drone-based cellular inspection. Background Technology

[0002] Rail transit projects involve long-distance lines and complex scenarios, creating an urgent need for real-time control and precise decision-making in construction and operation. The combination of digital twin technology and drone inspection provides an effective solution to the problems of low efficiency and scattered data in traditional manual control. The core is to build a virtual-physical twin by integrating multi-source data to support inspection optimization and dynamic on-site control.

[0003] Related technologies have been initially applied: the integration of BIM and GIS has enabled engineering visualization, and drone inspections have improved the efficiency of on-site data collection. At the same time, there have been preliminary practices in simple task scheduling, path planning methods, and multi-source data integration.

[0004] The core drawback of existing technologies lies in the failure to achieve deep correlation and dynamic adaptation of multi-source data, the disconnect between the twin model and the real-time status of the engineering site, the lack of flexible response to dynamic constraints in inspection task scheduling and path planning, and the insufficient matching degree between the distribution of drone swarms and task requirements. Ultimately, the accuracy and timeliness of engineering control cannot meet the actual needs of complex scenarios in rail transit engineering. Summary of the Invention

[0005] In order to achieve real-time and precise control of rail transit engineering and improve core control efficiency through digital twin virtual-real mapping and closed-loop intelligent inspection by drones, this application provides a digital twin control method for rail transit engineering based on drone honeycomb inspection.

[0006] Firstly, this application provides a digital twin control method for rail transit engineering based on UAV cellular inspection, employing the following technical solution:

[0007] A digital twin control method for rail transit engineering based on drone-based cellular inspection includes:

[0008] Acquire multi-source data for rail transit engineering, including BIM design models, GIS geographic information, IoT sensor data, and initial UAV imagery;

[0009] Extract the core parameters of the BIM design model, perform lightweight processing based on these parameters, and integrate multi-scale LOD rendering technology to adapt to scene requirements. Complete the registration with GIS geographic information through a preset coordinate system fusion algorithm, and establish a dynamic association network of engineering entities, environment and events by combining spatiotemporal tagging technology to form a digital twin benchmark model.

[0010] Based on real-time status data from a digital twin benchmark model, external factors are integrated, and a pre-set multi-dimensional evaluation algorithm is used to generate inspection task priorities. The real-time status data includes equipment health values ​​and environmental monitoring data, while the external factors include real-time weather, temporary construction areas, and train running times.

[0011] Based on the preset performance thresholds of the drone hive, a task-hive dynamic adaptation rule is established according to the priority of the inspection task. Drones are scheduled to perform inspections by avoiding restricted areas according to the preset path optimization algorithm. The performance thresholds include wind resistance level, endurance margin, and load capacity.

[0012] The above-mentioned inspections obtain multi-dimensional real-time data, which is then preprocessed at the edge and uploaded to the digital twin platform. The pre-set spatiotemporal data assimilation engine merges the data with the digital twin benchmark model, driving the model to iterate dynamically and achieve real-time synchronous updates between the twin and the engineering site.

[0013] By adopting the above technical solutions, a digital twin benchmark model is constructed by integrating multi-source data such as BIM and GIS. Inspection priorities are generated by combining multi-dimensional evaluation. Through the dynamic adaptation and obstacle avoidance inspection of drones, data feedback drives the real-time iteration of the model, breaking down data fragmentation and realizing real-time perception and precise control of the entire rail transit project, which greatly improves the efficiency of control. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall process of the digital twin control method for rail transit engineering based on UAV honeycomb inspection in this application embodiment.

[0015] Figure 2 This is another embodiment of the present application, which is a flowchart of generating inspection task priorities based on real-time status data of a digital twin benchmark model, incorporating external factors, and using a preset multi-dimensional evaluation algorithm. Detailed Implementation

[0016] The present application will be further described in detail below with reference to the accompanying drawings.

[0017] Reference Figure 1 The present application discloses a digital twin control method for rail transit engineering based on drone-based cellular inspection, comprising:

[0018] Step S100: Obtain multi-source data for rail transit engineering, including BIM design model, GIS geographic information, IoT sensor data, and initial UAV imagery.

[0019] The data includes IoT sensor data for real-time monitoring of equipment operating status (such as temperature, pressure, and vibration) and environmental conditions (such as temperature, humidity, and light). Initial drone imagery involves drones automatically capturing images of the construction site along preset flight paths (planned using software such as DJI GO). The images undergo preprocessing such as distortion correction and stitching to supplement details from the BIM model and GIS data.

[0020] The necessary processes are described below: 1. Data Fusion: By matching spatial location and timestamps, the BIM design model, GIS geographic information, IoT sensor data, and initial UAV imagery are integrated into a unified data framework. 2. Lightweight Processing: The BIM model is lightweighted using LOD rendering technology to simplify model details according to scene requirements and improve rendering efficiency. 3. Coordinate Transformation: Through coordinate transformation algorithms such as Helmert transformation, the GIS geographic information and the BIM model are accurately registered.

[0021] 4. Data Preprocessing: Cleaning and filtering IoT sensor data, and distortion correction and stitching of drone images to ensure data accuracy and usability.

[0022] Step S200: Extract the core parameters of the BIM design model, perform lightweight processing based on these parameters, and integrate multi-scale LOD rendering technology to adapt to scene requirements. Complete the registration with GIS geographic information through a preset coordinate system fusion algorithm, and establish a dynamic association network of engineering entities, environment, and events by combining spatiotemporal tagging technology to form a digital twin benchmark model.

[0023] Lightweight processing reduces model data size and improves rendering efficiency by simplifying model geometry and reducing texture complexity. Multi-scale LOD (Level of Detail) rendering technology dynamically adjusts the model's level of detail based on viewing distance, displaying a low-precision model at a distance and a high-precision model up close, thus optimizing rendering performance.

[0024] The acquisition methods and necessary processes are as follows: First, acquire the BIM design model, GIS geographic information, IoT sensor data, and initial UAV imagery. The BIM design model can be obtained from the design unit in IFC or RVT format, or exported through a BIM platform (such as Revit or Bentley). GIS geographic information is obtained from a geographic information system platform (such as ArcGIS Online) or through satellite remote sensing image processing. IoT sensor data is collected through the API interface of an IoT platform (such as Alibaba Cloud IoT), including equipment health values ​​and environmental monitoring data. Initial UAV imagery is collected using a UAV along a preset flight path and then processed using image processing software (such as Pix4D) for distortion correction and stitching. Next, extract the core parameters of the BIM design model, such as component geometry, material properties, and spatial location. This process is implemented through the API functions of the BIM software to ensure the accuracy and completeness of the extracted data. The BIM model is then lightweighted using multi-scale LOD rendering technology, dynamically adjusting model details based on the viewing distance. Specifically, a low-precision model is used for distant components, while a high-precision model is used for nearby components. This process is achieved through geometric simplification and texture optimization, ensuring optimal rendering results from different perspectives while significantly reducing the amount of model data and improving rendering efficiency.

[0025] Then, using coordinate system fusion algorithms such as Helmert transformation, the local coordinate system of the BIM model is precisely aligned with the global coordinate system of the GIS geographic information. This process includes translation, rotation, and scaling operations to ensure accurate matching of the model in geographic space. The specific steps are as follows: 1. Translation: Move the origin of the BIM model to the corresponding position in the GIS coordinate system. 2. Rotation: Adjust the orientation of the BIM model to match the orientation of the GIS coordinate system. 3. Scaling: Adjust the size of the BIM model as needed to ensure its scale is consistent with the GIS geographic information.

[0026] Subsequently, spatiotemporal tagging technology is used to assign time and space labels to engineering entities, environments, and events, establishing a dynamic relational network. This process is achieved by assigning a unique timestamp and spatial location label to each engineering entity and event, ensuring real-time data updates and synchronization.

[0027] Finally, the lightweight BIM model is integrated with GIS geographic information and combined with a dynamic correlation network to form a digital twin benchmark model.

[0028] Step S300 involves generating inspection task priorities based on real-time status data from the digital twin baseline model, incorporating external factors, and employing a pre-defined multi-dimensional evaluation algorithm. The real-time status data includes equipment health values ​​and environmental monitoring data, while external factors include real-time weather, temporary construction areas, and train operating times. The specific process of step S300 can be found in steps S310 to S360, and will not be elaborated upon here.

[0029] Step S400 involves establishing a task-hive dynamic adaptation rule based on the preset performance thresholds of the drone cellular network and the inspection task priority, and then scheduling drones to perform inspections by avoiding restricted areas using a preset path optimization algorithm. The performance thresholds include wind resistance, remaining battery life, and load capacity. The specific process of establishing the task-hive dynamic adaptation rule based on the preset drone cellular performance thresholds and the inspection task priority can be found in steps S410 to S430, and the specific process of scheduling drones to perform inspections by avoiding restricted areas using a preset path optimization algorithm can be found in steps S4A0 to S4E0; these details will not be elaborated here.

[0030] Step S500: Multi-dimensional real-time data is obtained through the above inspection. After preprocessing at the edge, the data is uploaded to the digital twin platform. The preset spatiotemporal data assimilation engine merges the data with the digital twin benchmark model, driving the model to iterate dynamically and achieve real-time synchronous updates between the twin and the engineering site.

[0031] Among them, the spatiotemporal data assimilation engine is a technology used to fuse real-time data with digital twin benchmark models. It can integrate new data into the model based on time and space information to ensure the real-time performance and accuracy of the model.

[0032] The acquisition method and necessary processes are described below: First, when performing inspection tasks, the drone acquires multi-dimensional real-time data through various onboard sensors and cameras. This data includes high-definition images, video streams, and environmental sensor data. This data is preprocessed at the edge of the drone, through data compression and format conversion, ensuring efficiency and accuracy during transmission. For example, image data may be compressed to reduce file size while retaining key information; sensor data may be calibrated to eliminate noise. Next, the preprocessed data is uploaded to the digital twin platform via a wireless network. On the platform, a pre-defined spatiotemporal data assimilation engine receives this data and fuses it with the digital twin baseline model according to time and spatial labels. This process involves data alignment and calibration, ensuring that new data accurately reflects the corresponding location and time point in the model. For example, if the drone takes images at a specific location, these images will be mapped to the corresponding location in the digital twin model. Subsequently, the spatiotemporal data assimilation engine drives the digital twin model to dynamically iterate. This means that the model automatically updates based on new data to reflect the latest status of the engineering site. For example, if a drone detects an anomaly in equipment in a certain area, the digital twin model will immediately update the status of that area, marking it as requiring attention or repair. This process is automated, ensuring that the model can reflect changes in the physical entity in real time. Ultimately, through this series of processes, the digital twin achieves real-time synchronization with the engineering site. This not only improves operational efficiency but also enhances the real-time monitoring capabilities of the engineering site, enabling managers to make timely decisions and ensure the safe and stable operation of rail transit projects.

[0033] Reference Figure 2 Based on real-time status data from a digital twin benchmark model, and incorporating external factors, a pre-defined multi-dimensional evaluation algorithm is used to generate inspection task priorities, including:

[0034] Step S310: Extract real-time status data from the digital twin baseline model and synchronously associate it with the importance level of the associated equipment, historical failure frequency, and external factors.

[0035] The acquisition method and necessary process are described below:

[0036] Real-time status data, including equipment health values ​​and environmental monitoring data, is extracted from the digital twin baseline model. Equipment health values ​​are collected in real time via an IoT sensor network, covering information such as equipment operating parameters and fault alarms; environmental monitoring data, including temperature, humidity, and air quality, is provided by environmental sensors distributed throughout the project site. This data is transmitted to the digital twin system via an IoT platform to ensure real-time updates. For example, for rail equipment, its health values ​​might include parameters such as track smoothness and wear, which are collected and transmitted in real time by sensors installed on the track.

[0037] The system synchronously correlates the importance level of equipment, historical failure frequency, and external factors. Equipment importance level is determined based on its criticality within the rail transit system and is used to assess the impact of equipment failure on the overall system. Historical failure frequency refers to the number of times the equipment has failed over a past period, used to assess equipment reliability. External factors are obtained through methods including real-time weather information from meteorological data service interfaces, temporary construction area information from the construction management system, and train operation schedule information from the rail transit dispatching system. This synchronous correlation of data is achieved through pre-defined rules and models, ensuring that all relevant data can be analyzed and processed within a unified framework. For example, for a critical piece of equipment with a high importance level and low historical failure frequency, but operating under adverse weather conditions, this data will be comprehensively considered to provide comprehensive data support for subsequent inspection task priority assessment.

[0038] Step S320: Normalize the real-time status data, equipment importance level, historical fault frequency and external factors, eliminate the difference in units by using a preset standardization algorithm, and remove data that deviates from the reasonable range by using a preset outlier filtering algorithm to form a standardized dataset.

[0039] The acquisition method and necessary process are described below:

[0040] First, real-time status data, equipment importance levels, historical failure frequencies, and external factors are normalized. This process is achieved through a pre-defined standardization algorithm, such as the Min-Max standardization method, which transforms all data to the range of 0 to 1. Specifically, for equipment health values, if their original range is 0 to 100, after Min-Max standardization, all values ​​will be transformed to the range of 0 to 1. The purpose of this is to eliminate the dimensional differences between different data points, enabling them to be compared and analyzed on the same scale. Next, a pre-defined outlier filtering algorithm is used to remove data that deviates from a reasonable range. For example, the 3σ principle based on statistical methods is used to identify and remove data points that exceed the normal range. For historical failure frequencies, if the historical failure frequency of a certain device is much higher than that of other similar devices, it may be an outlier and is removed using an outlier filtering algorithm. This ensures the accuracy and reliability of the data and avoids the impact of outlier data on subsequent analysis. Finally, a standardized dataset is formed. This dataset contains real-time status data, equipment importance levels, historical failure frequencies, and external factors after normalization and outlier filtering. For example, for a device, its health value after normalization is 0.8, its importance level is 0.9, its historical failure frequency is 0.2, and its external factor influence is 0.5.

[0041] Step S330: Based on the standardized dataset, a preset dynamic weighting algorithm is used to allocate weights for each dimension. The equipment importance level is coupled with the equipment health value in the real-time status data as the core weight. The weight of historical fault frequency is adjusted according to the time decay coefficient. External factors are given correction weights according to their degree of influence.

[0042] The acquisition method and necessary process are described below:

[0043] First, based on a standardized dataset, a pre-defined dynamic weighting algorithm is used to assign weights to each dimension. The device importance level and the device health value in the real-time status data are coupled as core weights. For example, for a critical device with a high importance level and a good current health value, these two indicators will be assigned higher core weights. Specifically, if the device importance level is 0.9 and the current health value is 0.8, through the dynamic weighting algorithm, the weights of these two indicators might be set to 0.45 and 0.4, totaling 0.85 of the core weight.

[0044] Next, the weight of historical failure frequencies is adjusted according to a time decay factor. This means that the impact of historical failure frequencies on the current state gradually decreases over time. For example, if a device has a high failure frequency in the past month, its impact will gradually weaken over time. The weight of historical failure frequencies can be dynamically adjusted using the time decay factor. Assuming an initial weight of 0.3 and a time decay factor of 0.9, the weight may be adjusted to 0.27 after one month.

[0045] Finally, external factors are assigned correction weights based on their degree of impact. For example, real-time weather conditions have a significant impact on equipment operation, and severe weather may increase the risk of equipment failure. Therefore, external factors are assigned corresponding correction weights based on the severity of real-time weather conditions. If the current weather conditions are severe, the correction weight might be set to 0.2 to reflect its potential impact on equipment operation.

[0046] Step S340: Input the weighted data of each dimension into the preset spatiotemporal correlation model, calculate the fault transmission correlation degree of adjacent equipment in combination with the spatial distribution characteristics of engineering entities, and generate an evaluation matrix containing spatiotemporal influence factors.

[0047] The spatiotemporal correlation model is a model that considers both temporal and spatial factors to assess the propagation correlation of equipment failures. Failure propagation correlation measures the degree of impact of a equipment failure on adjacent equipment, considering both temporal and spatial factors. The spatiotemporal impact factor is a quantitative indicator that integrates the impact of temporal and spatial factors on equipment failures. The evaluation matrix is ​​a matrix structure used to store and process multi-dimensional data, including information such as equipment health status, importance level, and historical failure frequency. The specific process can be found in steps S341 to S344, and will not be elaborated here.

[0048] Step S350 involves integrating the feature values ​​of the evaluation matrix data and the standardized dataset using a preset fusion algorithm to calculate the comprehensive priority score for each potential inspection task. The specific process can be found in steps S351 to S356.

[0049] Step S360: Sort the inspection tasks from high to low according to their comprehensive priority scores to generate a priority sequence.

[0050] The acquisition method and necessary process are described below: First, all potential inspection tasks are sorted according to their comprehensive priority scores. This process is implemented through a preset sorting algorithm to ensure that the task list is generated efficiently and accurately. For example, using the quicksort algorithm, inspection tasks can be quickly sorted from high to low according to their comprehensive priority scores. The purpose of this is to ensure that high-priority tasks are processed first, thereby improving inspection efficiency and response speed. Next, an inspection task priority sequence is generated. This sequence is an ordered task list in which the priority of each task is clearly defined. For example, suppose there are three inspection tasks in the system with comprehensive priority scores of 0.85, 0.70, and 0.90, respectively. After processing by the sorting algorithm, the generated inspection task priority sequence will be arranged in the order of 0.90, 0.85, and 0.70.

[0051] Generating an evaluation matrix that includes spatiotemporal influence factors includes:

[0052] Step S341: Input the weighted data of each dimension into the preset spatiotemporal correlation model. This model is based on the spatial distribution characteristics of engineering entities and constructs a spatial topology network of engineering entities through a built-in spatial topology algorithm, identifies adjacent devices in the network and marks the physical connection type.

[0053] The necessary process is described below:

[0054] First, the weighted data for each dimension is input into a pre-defined spatiotemporal correlation model. The core function of this model is to construct a spatial topology network of engineering entities using a built-in spatial topology algorithm. For example, in a rail transit system, engineering entities such as signaling equipment, track equipment, and communication equipment have specific spatial distributions and connections. The spatial topology algorithm can use adjacency matrices or adjacency lists from graph theory to represent the connections between these devices. Specifically, each device can be considered a node in the graph, and the connections between devices are considered edges. In this way, a complete spatial topology network can be constructed, providing a foundation for subsequent fault propagation analysis. Next, the spatiotemporal correlation model identifies adjacent devices in the network and labels their physical connection types. Physical connection types include direct connections and indirect connections, which directly affect the baseline value of the fault propagation probability. For example, the fault propagation probability between two directly connected signaling devices may be higher than that between indirectly connected devices. The process of labeling physical connection types can be achieved by analyzing the physical interfaces and communication protocols between devices. For example, if two devices are directly connected via optical fiber, this connection type can be labeled as "direct connection," while if two devices are connected via multiple intermediate devices, it can be labeled as "indirect connection." In this way, the model can more accurately assess the likelihood of fault propagation.

[0055] Step S342: The spatiotemporal correlation model, based on the labeled physical connection type, calls the built-in time correlation algorithm, introduces a preset time decay rule, and combines weighted historical fault frequency data and historical fault propagation records to calculate the probability of fault propagation from the source device to adjacent devices within different time periods. The physical connection type directly affects the baseline value of the propagation probability. Propagation probability: the probability of a fault propagating from one device to another, affected by the physical connection type and time factors. Historical fault propagation records: data recording the historical propagation path and impact range of device faults.

[0056] The acquisition method and necessary process are described below:

[0057] First, the spatiotemporal correlation model, based on the labeled physical connection types, invokes a built-in temporal correlation algorithm. This algorithm analyzes the propagation of faults at different points in time, combining weighted historical fault frequency data and historical fault propagation records to calculate the fault propagation probability. For example, assuming two devices A and B are directly connected, and historical data shows that in the past year, device B was affected by 70% of device A's faults. Therefore, the initial propagation probability can be set to 0.7.

[0058] Next, a pre-defined time decay rule is introduced to adjust the propagation probability. The time decay rule is typically expressed as a mathematical function, such as the exponential decay function. Where λ is the attenuation coefficient and t is time. Assume that after a fault occurs, its impact attenuates over time with a coefficient λ of 0.1. If device A failed 5 hours ago, the probability of the fault propagating to device B at the current time will be adjusted according to the time attenuation rule. The specific calculation is as follows: ;

[0059] Therefore, after 5 hours, the probability of a fault propagating from device A to device B decreased from an initial 0.7 to approximately 0.4246.

[0060] Finally, by combining weighted historical fault frequency data and historical fault propagation records, the probability of fault propagation from the source device to adjacent devices within different time periods is calculated. Assuming device A experienced 5 faults in the past month, and each fault affected adjacent device B, the historical fault frequency data would be given a higher weight. Simultaneously, the historical fault propagation records show that these faults propagated to adjacent device B within a specific timeframe; this data is used to adjust the baseline value of the propagation probability. The physical connection type directly affects the baseline value of the propagation probability; for example, the baseline value for directly connected devices might be 0.7, while the baseline value for indirectly connected devices might be 0.3. By integrating these factors, the model can more accurately calculate the fault propagation probability.

[0061] For example, suppose device A and device B are directly connected, and the baseline propagation probability is 0.7. Device A experienced 5 failures in the past month, each of which affected device B. The weight of the historical failure frequency data is 0.8. Combining the time decay rule and historical data, the final propagation probability can be calculated as follows: ;

[0062] Therefore, taking into account the physical connection type, time decay rules, and historical fault frequency data, the final probability of a fault propagating from device A to device B is approximately 0.3397.

[0063] Step S343: The spatiotemporal correlation model combines the identified physical connection type and the calculated transmission probability, and uses the built-in correlation fusion algorithm to fuse the spatial connection strength and temporal transmission characteristics to obtain the fault transmission correlation degree of adjacent devices.

[0064] The acquisition method and necessary process are described below:

[0065] In step S343, the spatiotemporal correlation model uses a correlation fusion algorithm to comprehensively evaluate the spatial connectivity strength and temporal transmission characteristics to calculate the fault transmission correlation degree. This process involves several technical details to ensure the accuracy and reliability of the calculation results.

[0066] First, the model determines the spatial connection strength based on the physical connection type. For example, if device A and device B are directly connected, their spatial connection strength baseline value is 0.8. If device C and device D are indirectly connected, their spatial connection strength baseline value might be 0.3. These baseline values ​​are pre-set based on the physical connection type between the devices; direct connections typically have higher spatial connection strength, while indirect connections have lower values. Next, the model further refines the calculation of fault propagation correlation by combining the temporal propagation probability calculated in step S342. The temporal propagation probability reflects the propagation characteristics of the fault in the time dimension and is calculated using the time decay rule. For example, if device A failed 5 hours ago, its propagation probability after time decay is 0.4246. This value reflects the probability of the fault propagating at the current time point. Then, the correlation fusion algorithm comprehensively evaluates the spatial connection strength and temporal propagation characteristics. Specifically, the algorithm uses a weighted summation method to fuse the spatial connection strength and temporal propagation probability into a comprehensive index. Assuming the weight of spatial connection strength is 0.6 and the weight of temporal propagation characteristics is 0.4, the calculation formula is as follows:

[0067] This value reflects the overall impact of a failure in device A on device B, taking into account both spatial and temporal factors.

[0068] Step S344: The spatiotemporal correlation model uses each device in the spatial topology network as the matrix row / column index, and uses the calculated fault propagation correlation degree as the matrix base value. It then overlays a preset time decay factor and a preset device importance correction coefficient to generate an evaluation matrix containing spatiotemporal influence factors.

[0069] The evaluation matrix is ​​a data structure used to store and process the fault propagation correlation between devices and related correction factors, providing a foundation for subsequent analysis. The device importance correction coefficient is a coefficient set according to the importance of the device in the system, used to adjust the fault propagation correlation.

[0070] In step S344, the spatiotemporal correlation model constructs an evaluation matrix based on the location of each device in the spatial topology network, using the devices as row and column indices. The model uses the fault propagation correlation degree calculated in step S343 as the base value of the matrix, and further superimposes the time decay factor and the device importance correction coefficient to generate an evaluation matrix that includes spatiotemporal influence factors.

[0071] The specific process is as follows:

[0072] Initialize the evaluation matrix: Based on the number of devices in the spatial topology network, create an n×n matrix, where n is the number of devices. The rows and columns of the matrix correspond to different devices.

[0073] Fill in the base values: Fill the corresponding positions in the matrix with the fault propagation correlation calculated in step S343. For example, if the fault propagation correlation between device A and device B is 0.65, then the positions (A,B) and (B,A) in the matrix will be filled with 0.65.

[0074] Superimposed Time Decay Factor: Based on a preset time decay rule, the time decay factor for each device pair is calculated and applied to the corresponding position in the matrix. For example, if the time decay factor between device A and device B is 0.9, then the values ​​of (A,B) and (B,A) in the matrix will be adjusted to 0.65 × 0.9 = 0.585.

[0075] Overlaying Equipment Importance Correction Factors: Based on the importance of the equipment, a correction factor is assigned to each equipment pair and applied to the corresponding position in the matrix. For example, if the importance correction factor for equipment A is 1.2 and the importance correction factor for equipment B is 1.1, then the value of (A,B) in the matrix will be adjusted to 0.585 × 1.2 × 1.1 = 0.7722.

[0076] The calculation of the overall priority score for each potential inspection task includes:

[0077] Step S351: Extract the feature values ​​of the assessment matrix data containing spatiotemporal influence factors and the standardized dataset, wherein the assessment matrix data characterizes the intensity of spatiotemporal correlation risk between devices.

[0078] The specific process is as follows: 1. Extract evaluation matrix data: Extract the spatiotemporal impact factors for each device pair from the evaluation matrix. These factors reflect the intensity of fault propagation risk between devices and are calculated based on the physical connection type, temporal propagation characteristics, and device importance correction coefficient. For example, the spatiotemporal impact factor between device A and device B might be 0.7722, indicating the degree of impact of a fault in device A on device B. 2. Extract feature values ​​from the standardized dataset: Extract key feature values ​​for each device from the standardized dataset, including device health value, importance level, and historical fault frequency. These feature values ​​are normalized to eliminate differences between different units, facilitating subsequent comprehensive evaluation. For example, device A's health value might be 0.85, its importance level 0.9, and its historical fault frequency 0.2. 3. Integrate data: Integrate the extracted evaluation matrix data and the feature values ​​from the standardized dataset into a unified data structure, providing a foundation for subsequent weight allocation and conflict resolution. For example, for device A, its integrated data may include a spatiotemporal impact factor of 0.7722, a health value of 0.85, an importance level of 0.9, and a historical failure frequency of 0.2.

[0079] Step S352: Based on the characteristics of rail transit engineering, a preset weight allocation rule is adopted to assign preset weights to the equipment health and importance coupling value, historical fault and external factor correction value in the evaluation matrix data and the feature values ​​of the standardized dataset.

[0080] The specific process is as follows: 1. Determine the weight allocation rules: Based on the characteristics and needs of rail transit engineering, preset weight allocation rules for different dimensions of data. For example, the equipment health and importance coupling value may be assigned a higher weight because these indicators directly reflect the current state of the equipment and its importance to the system. Historical failure frequency and external factor correction value may also be assigned certain weights to reflect the historical performance of the equipment and the impact of the current external environment. 2. Allocate weights: According to the preset weight allocation rules, assign weights to the evaluation matrix data, equipment health and importance coupling value, and historical failure and external factor correction value respectively. For example, assume the weight of the evaluation matrix data is 0.4, the weight of the equipment health and importance coupling value is 0.3, the weight of the historical failure frequency is 0.2, and the weight of the external factor correction value is 0.1. 3. Calculate the base score for each dimension: Multiply the data of each dimension by its corresponding weight to obtain the base score for each dimension. For example, for equipment A, its evaluation matrix data is 0.7722, its equipment health value is 0.85, its importance level is 0.9, its historical failure frequency is 0.2, and its external factor correction value is 0.1.

[0081] The base scores for each dimension are calculated as follows:

[0082] The basic score for the evaluation matrix data is 0.7722 × 0.4 = 0.30888; the basic score for the coupling value of equipment health and importance is (0.85 × 0.9) × 0.3 = 0.2295; the basic score for historical failure frequency is 0.2 × 0.2 = 0.04; and the basic score for external factor correction value is 0.1 × 0.1 = 0.01.

[0083] Integrating Base Scores: The base scores from each dimension are integrated into a unified data structure, providing a foundation for subsequent comprehensive priority score calculations. For example, the integrated base score for device A is:

[0084] Basic score for the assessment matrix data: 0.30888. Basic score for the coupling value between equipment health and importance: 0.2295. Basic score for historical failure frequency: 0.04. Basic score for external factor correction value: 0.01.

[0085] Step S353: Based on the extracted evaluation matrix data and the feature values ​​of the standardized dataset, a preset conflict resolution algorithm is used to handle potential contradictions between the data, ensuring the consistency of data from different dimensions during fusion.

[0086] The specific process is as follows: 1. Identify potential contradictions: First, by comparing the feature values ​​of the evaluation matrix data and the standardized dataset, potential contradictions are identified. For example, the evaluation matrix shows a high fault transmission correlation between device A and device B (0.7722), but the standardized dataset shows a high health value for device B (0.9). This may indicate that although device B has a high fault transmission correlation with device A, its own condition is good. The contradiction lies in the mismatch between high correlation and high health value. 2. Apply conflict resolution algorithms: The preset conflict resolution algorithm can use various methods to handle these contradictions. A common method is rule-based adjustment. For example, if a device's health value is higher than a certain threshold (e.g., 0.8), even if its fault transmission correlation is high, its correlation weight can be appropriately reduced. Another method is to use statistical methods, such as taking the average or median, to smooth the differences between data. Rule-based adjustment: If device B's health value is higher than 0.8, the correlation weight between device A and device B in the evaluation matrix can be reduced. For example, the correlation can be adjusted from 0.7722 to 0.6. Statistical methods: If there are inconsistencies among multiple devices, statistical methods can be used to smooth the data. For example, the average health value and average correlation degree of multiple devices can be taken to reduce the impact of extreme values ​​of individual devices on the overall data. 3. Adjusting the data: Based on the results of the conflict resolution algorithm, the data can be adjusted. For example, if it is decided to reduce the fault propagation correlation degree weight of device B, the correlation degree between device A and device B in the evaluation matrix can be adjusted from 0.7722 to 0.6. At the same time, the health value of device B can also be fine-tuned to better reflect its actual status.

[0087] 4. Verify Consistency: Finally, verify whether the adjusted data achieves consistency. This can be done by checking whether the adjusted data conforms to preset logical rules and numerical ranges. For example, ensure that the health values ​​and fault propagation correlations of all devices are within reasonable ranges and that there are no obvious contradictions between them. Specific verification methods may include:

[0088] Logical check: Ensure the adjusted data is logically sound; for example, a device's health value should not exceed 1. Numerical range check: Ensure all data values ​​are within a preset range; for example, correlation should be between 0 and 1. Consistency check: Ensure the adjusted data does not exhibit significant contradictions across different dimensions; for example, devices with high health values ​​should not have high fault propagation correlation.

[0089] Step S354: Map the conflict-resolved evaluation matrix data to a single device dimension according to the device correspondence, and multiply the standardized dataset feature values ​​by the corresponding preset fusion dimension weights to obtain the basic scores for each dimension.

[0090] The specific process is as follows: 1. Mapping the evaluation matrix data to the single-device dimension: First, map the data in the evaluation matrix to the single-device dimension according to the device correspondence. For example, the fault propagation correlation degree between device A and device B is C. AB =0.7722, the fault propagation correlation degree between equipment A and equipment C is C AC =0.5. These correlations will be mapped to the single-device dimension of device A, denoted as S. A :S A =C AB +C AC =0.7722 + 0.5 = 1.2722. 2. Extract feature values ​​from the standardized dataset: Extract feature values ​​for device A from the standardized dataset, including the device health value H. A Importance Level I A Historical Fault Frequency F A For example, the health value H of device A. A =0.85, Importance Level I A =0.9, historical fault frequency F A =0.2. 3. Multiply by preset fusion dimension weights: Based on the preset fusion dimension weights, multiply the feature values ​​of the evaluation matrix data and the standardized dataset by their respective weights. Assume the preset weights are as follows:

[0091] Evaluation matrix data weights W S =0.4. Equipment health value weight W H =0.3. Importance level weight W I =0.2. Historical fault frequency weight W F =0.1. The base scores for each dimension are calculated as follows:

[0092] Evaluation matrix data base score S A ′=S A ×W S =1.2722 × 0.4 = 0.50888. Equipment health base score H A =H A ×W H =0.85 × 0.3 = 0.255. Base score for importance level I. A =I A ×W I =0.9 × 0.2 = 0.18. Historical fault frequency baseline score F A =F A ×W F =0.2×0.1=0.02.

[0093] 4. Integrate Base Scores: Integrate the base scores from each dimension into a unified data structure to provide a foundation for subsequent comprehensive priority score calculations. For example, the integrated base score for device A is:

[0094] Evaluation matrix data base score S A =0.50888. Equipment health base score H A =0.255. Base score I for importance level. A =0.18. Historical fault frequency baseline score F A =0.02.

[0095] Step S355: For the high-frequency risk areas and areas affected by external emergency factors that are centrally labeled in the standardized dataset, a preset dynamic adjustment factor is introduced to adjust the scores of the high-frequency risk area tasks and the tasks affected by external emergency factors according to the preset coefficients.

[0096] High-frequency risk areas: Areas marked in standardized datasets that have historically experienced frequent failures or high risks. Areas affected by external emergencies: Areas affected by external emergencies (such as severe weather, temporary construction, etc.).

[0097] The specific process is as follows: 1. Identify high-frequency risk areas and areas affected by external emergency factors: First, extract labeled high-frequency risk areas and areas affected by external emergency factors from the standardized dataset. For example, device A is located in a high-frequency risk area, and device B is located in an area affected by external emergency factors due to severe weather. 2. Introduce dynamic adjustment factors: According to preset rules, introduce dynamic adjustment factors for high-frequency risk areas and areas affected by external emergency factors. Assume the preset dynamic adjustment factors are as follows: Dynamic adjustment factor for high-frequency risk areas α = 1.2. Dynamic adjustment factor for areas affected by external emergency factors β = 1.3. 3. Adjust task scores: Multiply the tasks located in high-frequency risk areas and areas affected by external emergency factors by the corresponding dynamic adjustment factors. For example, the base score of device A is S. A =0.50888, the base score of device B is S B =0.45. The adjusted score is as follows:

[0098] The formula for calculating the adjusted score for Equipment A (high-frequency risk area) is as follows:

[0099] .

[0100] The formula for calculating the adjusted score for Equipment B (area affected by external emergency factors) is as follows:

[0101] .

[0102] Step S356: The basic scores of each dimension are weighted and summed with the corresponding dynamic adjustment factors to obtain the comprehensive priority score of each potential inspection task.

[0103] Based on the preset performance thresholds of the drone cellular network, and according to the priority of inspection tasks, dynamic adaptation rules for task-cellular networks are established, including:

[0104] Step S410: Extract the preset performance threshold parameters of the drone hive, the real-time status data of the drone, and the key parameters of the inspection task. The key parameters of the inspection task include the inspection area, the type of target equipment and the required detection accuracy, and the timeliness requirements corresponding to the task priority.

[0105] The specific process is as follows: 1. Extract performance threshold parameters: Extract preset performance threshold parameters from the UAV cellular management system, including wind resistance level, remaining flight time, and payload capacity. These parameters are the basic performance indicators of the UAV cellular system, used to evaluate whether the cellular system can meet the needs of the inspection task. For example, the wind resistance level is level 5, the remaining flight time is 30 minutes, and the payload capacity is 2 kg. 2. Extract real-time UAV status data: Obtain real-time status data of the UAV from the UAV status monitoring system, including current battery level, flight status, and payload status. This data reflects the current availability and operating status of the UAV. For example, the current battery level is 80%, the flight status is standby, and the payload status is no load. 3. Extract key parameters for the inspection task: Extract key parameters for the inspection task from the inspection task management system, including the inspection area, target equipment type, required detection accuracy, and timeliness requirements corresponding to the task priority. These parameters define the specific requirements and priorities of the inspection task. For example, the inspection area is 10 square kilometers, the target equipment type is a track signal device, the required detection accuracy is centimeter-level, and the timeliness requirement corresponding to the task priority is completion within 2 hours.

[0106] Step S420: Synchronously acquire external factor data as constraints for the adaptation rules. External factor data includes wind speed and precipitation levels from real-time meteorological data, real-time boundary update data of temporary construction areas, and no-fly periods information from train operation plans.

[0107] Among them, the adaptation rule constraints are the restrictions that must be met during the task-honeycomb dynamic adaptation process to ensure the safety and effectiveness of the inspection task.

[0108] The specific process is as follows: 1. Obtain real-time meteorological data: Obtain real-time meteorological data, including wind speed and precipitation level, from the meteorological service interface. This data is used to assess whether the current weather conditions are suitable for drone flight. For example, if the wind speed exceeds the drone's wind resistance threshold or the precipitation level is too high, it may affect the drone's stability and mission execution effectiveness. 2. Obtain temporary construction area boundary update data: Obtain real-time boundary update data of the temporary construction area from the construction management system. This data helps avoid construction areas, preventing conflicts between the drone and construction equipment or personnel, and ensuring the safety of the inspection mission. 3. Obtain no-fly periods information from train operation plans: Obtain no-fly periods information from the train dispatching system from the train operation plan. This information is used to plan the drone's inspection time, avoiding inspections when trains are passing at high speeds, ensuring the safety and accuracy of the inspection mission.

[0109] Step S430: Based on the inspection task priority, combined with the aforementioned performance threshold parameters, key task parameters and external factor data, a dynamic adaptation rule is established through a preset priority-performance matching logic. Among them, high-priority tasks are matched with the core performance standard first, low-priority tasks are matched according to the resource balancing principle, and sudden factors trigger the switching of the backup cell for high-priority tasks.

[0110] The dynamic adaptation rules are as follows: Rules that dynamically match inspection tasks with drone cellular resources based on task priority and drone cellular performance. Priority-performance matching logic: Predefined logic used to match appropriate drone cellular resources according to the priority of the inspection task. Core performance compliance: The drone cellular performance meets the key parameter requirements of the inspection task, such as wind resistance, remaining battery life, and load capacity. Resource balancing principle: In low-priority tasks, drone cellular resources are evenly allocated based on resource usage to avoid excessive resource concentration. Backup cellular switching: When unforeseen factors affect high-priority tasks, backup cellular resources are switched to ensure task continuity and safety.

[0111] The specific process is as follows:

[0112] 1. High-priority task matching:

[0113] Extracting Task Priorities: Extracting task priority information from the inspection task management system. Filtering Hives Meeting Core Performance Standards: Based on key task parameters (such as inspection area range, target equipment type, required detection accuracy, etc.), filtering Hives meeting core performance standards from the drone swarm management system. For example, if a task requires a swarm with wind resistance level 5, 30 minutes of remaining flight time, and a payload capacity of 2 kg, the system will filter out swarm resources that meet these conditions. Allocating Swarm Resources: Prioritizing the allocation of filtered swarm resources to high-priority tasks. For example, the system will allocate swarms with wind resistance level 5, 30 minutes of remaining flight time, and a payload capacity of 2 kg to high-priority tasks.

[0114] 2. Low-priority task matching:

[0115] Extract Remaining Cellular Resources: After high-priority tasks are assigned, extract the remaining cellular resources. Calculate Resource Load: Calculate the current load and availability of each remaining cellular unit. For example, calculate the current number of tasks and remaining power for each cellular unit. Balance Resource Allocation: Based on resource load and availability, evenly allocate cellular resources to low-priority tasks. For example, assign cellular units with lower loads to low-priority tasks to ensure balanced resource usage.

[0116] 3. Backup Hive Switching:

[0117] Real-time monitoring of cellular status: The system monitors the operational status of cellular drones in real time, including battery level, flight status, and payload. Detection of unexpected events: The system detects unexpected events such as sudden changes in weather conditions or equipment failures. For example, if a cellular drone currently performing a high-priority task becomes unusable due to a malfunction, the system will trigger a backup cellular switchover. Triggering backup cellular status: The system selects a suitable cellular drone from a pre-set backup cellular resource pool to ensure the continuity and safety of high-priority tasks. For example, the system assigns a backup cellular drone (such as one with wind resistance level 5, 30 minutes of remaining flight time, and a payload capacity of 2 kg) to an affected high-priority task.

[0118] Dispatching drones to perform inspections by using preset path optimization algorithms to avoid restricted areas includes:

[0119] Step S4A0: Based on the matching results generated by the task-honeycomb dynamic adaptation rules, obtain the spatiotemporal constraints of the restricted area from the digital twin benchmark model and external systems, including the coordinates and time periods of the no-fly zone for trains, the isolation boundary of the temporary construction area, and the range of the high-risk meteorological area, and transform them into preset spatial and temporal constraints for path planning.

[0120] The system includes: Task-Hospital Dynamic Adaptation Rules: Rules for dynamically matching tasks and cellular resources based on inspection task priority and UAV cellular performance. Restricted Area Spatiotemporal Constraints: These include coordinates and time periods of train no-fly zones, isolation boundaries of temporary construction areas, and the scope of high-risk weather zones, used to restrict the UAV's flight area and time. Path Planning Constraints: These transform the restricted area spatiotemporal constraints into input conditions for the path planning algorithm, ensuring that the generated path avoids these areas.

[0121] The specific process is as follows:

[0122] 1. Extract data from restricted areas:

[0123] Train No-Fly Zone Coordinates and Time Range: Train operation plans are obtained from the train dispatching system to extract the coordinates and time range of the no-fly zone. For example, if a train is passing through a section of track at high speed during a specific time period, the drone must avoid that area. Temporary Construction Zone Isolation Boundary: Real-time boundary update data of temporary construction zones is obtained from the construction management system. This data defines the scope of the construction area, and drones must maintain a safe distance. High-Risk Meteorological Zone Range: Real-time meteorological data is obtained from the meteorological service interface to extract the range of high-risk zones. For example, areas with strong winds or precipitation may affect the stability and safety of the drone, and these areas must be avoided.

[0124] 2. Transform into path planning constraints:

[0125] Spatial constraints: Transforming the geographical coordinates of the restricted area into spatial constraints for the path planning algorithm. For example, inputting the coordinates of no-fly zones and temporary construction zones into the path planning algorithm ensures that the generated path avoids these areas. Temporal constraints: Transforming the time range of the restricted area into temporal constraints for the path planning algorithm. For example, inputting the time periods during which trains pass into the path planning algorithm ensures that drones do not enter no-fly zones during these time periods.

[0126] 3. Integrate constraints:

[0127] Integrate spatial and temporal constraints: Integrate the extracted spatial and temporal constraints into a unified data structure as input conditions for the path planning algorithm. For example, create a constraint matrix containing spatial and temporal information for all restricted areas. Validate constraints: Ensure all constraints are logically and numerically consistent and conflict-free. For example, check if the time periods for train no-fly zones overlap with those for temporary construction zones to ensure drones do not enter any restricted areas during these periods.

[0128] Step S4B0: Extract the performance parameters and inspection task requirements of the matching UAV and input them into the preset path optimization algorithm. The preset path optimization algorithm prioritizes meeting the time requirements and minimizing energy consumption. Combining the aforementioned constraints and parameters, it generates an initial path that avoids all restricted areas, including the coordinates of the waypoints, flight speed, and estimated time nodes.

[0129] Among them, the drone performance parameters are: the technical specifications and capabilities of the drone, such as maximum flight speed, endurance, payload capacity, and wind resistance level. Inspection task requirements are: the specific requirements of the inspection task, including the inspection area, target equipment type, required detection accuracy, and timeliness requirements corresponding to task priority. The path optimization algorithm is an algorithm used to generate flight paths that meet specific conditions (such as timeliness requirements and minimum energy consumption) while avoiding restricted areas.

[0130] The specific process is as follows:

[0131] 1. Extract UAV performance parameters:

[0132] The drone management system retrieves and extracts performance parameters such as maximum flight speed, flight time, payload capacity, and wind resistance rating of the matched drones. These parameters define the drone's flight capabilities and limitations. For example, a certain model of drone has a maximum flight speed of 20 meters per second, a flight time of 30 minutes, a payload capacity of 2 kilograms, and a wind resistance rating of level 5.

[0133] 2. Extract inspection task requirements:

[0134] Obtain key parameters from the inspection task management system, including the inspection area, target equipment type, required detection accuracy, and timeliness requirements corresponding to the task priority. These parameters define the specific requirements of the inspection task. For example, an inspection task may require covering an area of ​​10 square kilometers, targeting track signal equipment, requiring centimeter-level detection accuracy, and requiring completion within 2 hours.

[0135] 3. Input to path optimization algorithm:

[0136] Define the optimization objective: The objective of the path optimization algorithm is to meet timeliness requirements while minimizing energy consumption. This means that the generated path should complete the inspection task within the specified time while minimizing the drone's energy consumption.

[0137] Combining constraints and parameters: The extracted UAV performance parameters and inspection task requirements are input into the path optimization algorithm, along with the spatiotemporal constraints of the restricted areas obtained in step S4A0 (such as the coordinates and time periods of train no-fly zones, the isolation boundaries of temporary construction zones, and the range of high-risk meteorological areas). These constraints and parameters together guide the algorithm to generate an initial path that avoids all restricted areas.

[0138] Initial Path Generation: The path optimization algorithm generates an initial path based on the input parameters and constraints, including the coordinates of waypoints, flight speed, and estimated time points. For example, the generated path might include the following key points: Start Point: UAV hive location. Waypoint 1: Coordinates (40.123456, 116.123456), flight speed 15 m / s, estimated arrival time 08:30:00. Waypoint 2: Coordinates (40.123457, 116.123457), flight speed 10 m / s, estimated arrival time 08:35:00. End Point: End point of the inspection area, coordinates (40.123458, 116.123458), flight speed 15 m / s, estimated arrival time 08:40:00.

[0139] 4. Verify the feasibility of the path:

[0140] Check if the path avoids restricted areas: Verify that the generated initial path completely avoids all restricted areas, such as no-fly zones for trains, temporary construction zones, and high-risk weather zones.

[0141] Check if the route meets the timeliness requirements: Verify that the total flight time of the route is within the timeliness requirements of the inspection task.

[0142] Check if the path meets the drone's performance parameters: Verify that the flight speed and endurance requirements of the path are within the drone's performance range.

[0143] In step S4C0, the initial path is simulated for compliance using a digital twin 3D scene. If there is overlap with the restricted area or insufficient safety distance, the algorithm is triggered to iterate and optimize again until the path fully complies with the constraints.

[0144] Compliance simulation involves simulating the initial path in a digital twin 3D scene to check if the path meets safety and operational requirements. Restricted areas are areas where drones are prohibited from entering or where a safe distance must be maintained, such as no-fly zones for trains, temporary construction zones, and high-risk weather zones. Safe distance is the minimum distance that a drone must maintain between itself and the restricted area to ensure flight safety.

[0145] The specific process is as follows:

[0146] 1. Import initial path and restricted area data:

[0147] Initial path data: Initial path data is obtained from the path optimization algorithm, including the coordinates of waypoints, flight speed, and estimated time points.

[0148] Restricted area data: Obtain the geometry and location information of restricted areas from digital twin baseline models and external systems, including no-fly zones for trains, temporary construction zones, and high-risk meteorological zones.

[0149] 2. Simulation in a digital twin 3D scene:

[0150] Constructing a 3D scene: Constructing a 3D scene on a digital twin platform, including the initial position of the drone, the inspection area, the position of the target equipment, and the restricted area.

[0151] Simulated flight path: Input the initial path data into the 3D scene to simulate the drone's flight process. For example, simulate the drone starting from the starting point, passing through waypoint 1 and waypoint 2 in sequence, and finally reaching the destination.

[0152] Compliance checks: During simulation, check whether the path overlaps with restricted areas or the safety distance is insufficient. For example, check whether the drone enters a train no-fly zone or the distance to a temporary construction area is less than the safety distance during flight.

[0153] 3. Trigger a new iteration and optimization:

[0154] Detect violation points: If the simulation results show that the path overlaps with the restricted area or the safety distance is insufficient, record the location and time of the violation point.

[0155] Path adjustment: Feedback on violation points to the path optimization algorithm, triggering a new iteration for optimization. For example, adjusting the coordinates of waypoints or flight speed to avoid restricted areas.

[0156] Repeated simulation: Perform compliance simulation again on the adjusted path to ensure that the path fully complies with the constraints.

[0157] 4. Verify the final path:

[0158] Final path check: After multiple iterations of optimization, verify that the final path fully meets all constraints, including avoiding all restricted areas and meeting safety distance requirements.

[0159] Generate a compliant path: Once the path passes the compliance simulation, the final compliant path is generated, including the coordinates of the waypoints, the flight speed, and the estimated time points.

[0160] Step S4D0: Send execution instructions containing optimized paths and sensor operating modes to the matched drones, and schedule them to start inspections according to the paths.

[0161] The specific process is as follows:

[0162] 1. Generate execution instructions:

[0163] Integrated and optimized path: The path data, after compliance simulation and iterative optimization, is integrated into the execution command. The path data includes the coordinates of waypoints, flight speed, and estimated time points.

[0164] Integrated sensor operating modes: Based on the requirements of the inspection task, set the sensor operating modes, including data acquisition frequency, resolution, and operating time. For example, set the camera to acquire high-definition images at a frequency of 30 frames per second, and the LiDAR to acquire point cloud data at a frequency of 100 times per second.

[0165] Generate an instruction set: Integrate the optimized path and sensor operating modes into a single instruction set to form complete execution instructions. For example, the instruction set might include the following:

[0166] Path data: Coordinates of waypoint 1 (40.123456, 116.123456), flight speed 15m / s, estimated arrival time 08:30:00.

[0167] Sensor settings: Camera acquisition frequency 30fps, LiDAR acquisition frequency 100Hz, data storage path / data / collection / task_001.

[0168] 2. Issue execution instructions:

[0169] Establish a communication connection: Establish a communication connection with the matched drone through the drone control system to ensure that commands can be transmitted reliably.

[0170] Transmit command set: The generated execution commands are transmitted to the drone via a wireless communication module. For example, the command set can be sent to the drone's flight control system using Wi-Fi or 4G / 5G networks.

[0171] Confirm command reception: Verify whether the drone has successfully received and parsed the command set. For example, after receiving a command, the drone returns a confirmation message, acknowledging that the command has been correctly received and is ready for execution.

[0172] 3. Initiate the inspection task:

[0173] Initialize the drone: The drone initializes its flight control system and sensors based on the received instruction set. For example, it sets the flight speed, navigation path, and sensor parameters.

[0174] Inspection Start: The drone initiates the inspection mission according to the optimized path and begins data collection. For example, the drone starts from the starting point, passes through waypoint 1 and waypoint 2 in sequence, and finally reaches the destination, collecting data according to the preset sensor working mode.

[0175] Real-time monitoring: The drone's flight status and data acquisition are monitored in real time through the drone control system. For example, the drone's location, remaining battery power, flight speed, and sensor data quality are monitored.

[0176] Step S4E0: Receive real-time updates on the drone's location, remaining battery power, and external factors. If a conflict is detected between the path and the updated restricted area, immediately invoke the preset dynamic obstacle avoidance sub-algorithm to generate an alternative path.

[0177] The specific process described above can be found in steps S4E1 to S4E5, and will not be repeated here.

[0178] If a conflict is detected between the path and the updated restricted area, the preset dynamic obstacle avoidance sub-algorithm is immediately invoked to generate an alternative path, including:

[0179] Step S4E1: Receive the drone's location data and remaining battery information in real time, and simultaneously acquire dynamically updated data on external factors to form the basis data for conflict analysis.

[0180] The specific process is as follows:

[0181] 1. Receive drone data in real time:

[0182] Location data: The drone's location information, including latitude, longitude, and altitude, is obtained in real time through its GPS module. For example, the drone's current location is (40.123456, 116.123456, 10m).

[0183] Remaining battery information: The remaining battery level is obtained in real time through the drone's battery management system. For example, the remaining battery level is 60%.

[0184] 2. Synchronously acquire dynamically updated data on external factors:

[0185] Meteorological data: Real-time meteorological data, including wind speed and precipitation level, is obtained from the meteorological service interface. For example, the current wind speed is 5 m / s and the precipitation level is light.

[0186] Temporary construction zone boundary: Obtain real-time boundary update data for the temporary construction zone from the construction management system. For example, the construction zone boundary is updated to (40.123457, 116.123457, 5m).

[0187] No-fly zone: Obtain updated no-fly zone data from the air traffic control system. For example, add a new no-fly zone (40.123458, 116.123458, 10m).

[0188] Integrating basic data for conflict analysis:

[0189] Data fusion: Integrating the drone's location data, remaining battery information, and dynamically updated data from external factors into a unified data structure.

[0190] Step S4E2: Based on the conflict analysis data, calculate the spatial overlap and temporal conflict coefficient between the current planned path and the updated restricted area using a preset conflict determination model. When any indicator reaches the preset conflict threshold, trigger the obstacle avoidance response mechanism.

[0191] The conflict determination model is a preset model used to calculate the spatial overlap and temporal conflict coefficient between the current planned path and the updated restricted area. Spatial overlap: An indicator measuring the degree of spatial overlap between the current planned path and the restricted area. Temporal conflict coefficient: An indicator measuring the degree of temporal conflict between the current planned path and the restricted area. Conflict threshold: A preset threshold that triggers the obstacle avoidance response mechanism when the spatial overlap or temporal conflict coefficient reaches this value.

[0192] The specific process is as follows:

[0193] 1. Extract basic data for conflict analysis:

[0194] Drone location data: Extract current location information, including latitude, longitude, and altitude, from the data transmitted back by the drone in real time. For example, the drone's current location is (40.123456, 116.123456, 10m).

[0195] Remaining battery information: Extract the remaining battery level from the drone's battery management system. For example, the remaining battery level is 60%.

[0196] External factors are dynamically updated: Real-time updated restricted area data is extracted from external systems, including weather conditions, temporary construction zone boundaries, no-fly zones, etc. For example, a new no-fly zone is added (40.123458, 116.123458, 10m).

[0197] 2. Calculate spatial overlap:

[0198] Extract path and restricted area coordinates: Extract the coordinates of waypoints from the current planned path and extract the boundary coordinates from the restricted area data.

[0199] Calculate overlap: Use geometric algorithms to calculate the spatial overlap between the path and the restricted area. For example, use a polygon intersection algorithm to calculate the overlap area between the path and the no-fly zone. Assume the overlap area between the path and the no-fly zone is A. overlap The total area of ​​the no-fly zone is A. total Then the spatial overlap S is:

[0200] .

[0201] Example: If the overlap area between the flight path and the no-fly zone is 50m 2 The total area of ​​the no-fly zone is 100m. 2 If the spatial overlap is S=0.5.

[0202] 3. Calculate the time conflict coefficient:

[0203] Extract time information: Extract the estimated arrival time for each waypoint from the current planned path, and extract the time window from the restricted area data.

[0204] Calculate the conflict coefficient: Use a time window comparison algorithm to calculate the time conflict coefficient between the path and the restricted area. For example, calculate the overlap time between the path's expected arrival time and the time window of the restricted area. Assume the path's expected arrival time is T. path The time window for the restricted region is T. restricted Then the time conflict coefficient C is:

[0205] .

[0206] Example: If the estimated arrival time of the path is 10 minutes and the overlap time with the restricted area time window is 5 minutes, then the time conflict factor C = 0.5.

[0207] 4. Trigger obstacle avoidance response mechanism:

[0208] Set conflict threshold: Preset spatial overlap threshold S threshold and time conflict coefficient threshold C threshold For example, let S be set. threshold =0.3 and C threshold =0.3.

[0209] Conflict detection: If the spatial overlap S ≥ S threshold Or the time conflict coefficient C ≥ C threshold If the calculated spatial overlap S=0.5 or temporal conflict coefficient C=0.5 both exceed the preset threshold of 0.3, the obstacle avoidance response mechanism will be triggered. Example: If the calculated spatial overlap S=0.5 or temporal conflict coefficient C=0.5 both exceed the preset threshold of 0.3, the obstacle avoidance response mechanism will be triggered.

[0210] Step S4E3: After the obstacle avoidance response mechanism is triggered, extract the core parameters from the basic conflict analysis data, associate them with the remaining time requirement of the task and the meteorological energy consumption correction parameters, and use them as input data for the preset dynamic obstacle avoidance sub-algorithm.

[0211] The specific process is as follows:

[0212] 1. Extract core parameters:

[0213] Drone location data: Extract the current location information of the drone from the conflict analysis base data, including latitude, longitude, and altitude. For example, the drone's current location is (40.123456, 116.123456, 10m).

[0214] Remaining battery information: Extract the drone's remaining battery power from the conflict analysis baseline data. For example, the remaining battery power is 60%.

[0215] Restricted Area Data: Updated restricted area information is extracted from the conflict analysis baseline data, including location, extent, and time window. For example, a new no-fly zone is added (40.123458, 116.123458, 10m).

[0216] 2. Remaining time limit for associated tasks:

[0217] Extract Remaining Task Time: Retrieve the remaining time for the task from the inspection task management system. For example, the remaining time for the task is 30 minutes.

[0218] Calculate the time urgency factor: Based on the remaining mission time and current flight progress, calculate the time urgency factor. For example, the time urgency factor T... urgency For: T urgency = Remaining time / Estimated completion time.

[0219] Example: If the task has 30 minutes remaining and an estimated completion time of 60 minutes, then T urgency =0.5.

[0220] 3. Extract meteorological energy consumption correction parameters:

[0221] Real-time weather data: Obtain real-time weather data from the weather service interface, including wind speed and precipitation level. For example, the current wind speed is 5 m / s and the precipitation level is light.

[0222] Calculate energy consumption correction factors: Based on real-time meteorological data, calculate energy consumption correction factors. For example, wind speed correction factor W. correction For: W correction =1 + wind speed / 10. Example: If the current wind speed is 5 m / s, then W correction =1 + 5 / 10 = 1.5.

[0223] 4. Integrate input data:

[0224] Core parameters are integrated: The extracted UAV location data, remaining battery power information, restricted area data, remaining mission time requirements, and weather energy consumption correction parameters are integrated into a unified data structure as input data for the dynamic obstacle avoidance sub-algorithm.

[0225] Step S4E4: Invoke the preset dynamic obstacle avoidance sub-algorithm. With obstacle avoidance safety as the core objective and minimizing time loss, generate a preset number of candidate alternative paths based on the input parameters. All paths must meet the dual constraints of a safe distance from the restricted area ≥ a preset value and estimated energy consumption ≤ a preset proportion of remaining power.

[0226] The specific process is as follows:

[0227] 1. Call the dynamic obstacle avoidance sub-algorithm:

[0228] Input parameters: Input the input data (UAV location, remaining battery power, restricted area data, remaining mission time requirements, weather energy consumption correction parameters, etc.) integrated in step S4E3 into the dynamic obstacle avoidance sub-algorithm.

[0229] Algorithm objective: To generate multiple candidate alternative paths with the core objectives of prioritizing obstacle avoidance safety and minimizing time loss.

[0230] 2. Generate candidate alternative paths:

[0231] Path sampling: Multiple potential paths are generated between the UAV's current position and the target position using sampling methods. For example, the Fast Expanding Random Tree (RRT) algorithm or the artificial potential field method can be used to generate paths.

[0232] Path optimization: The generated path is optimized to ensure it is smooth and meets flight performance requirements. For example, Bézier curves or polynomial interpolation can be used to optimize the path shape.

[0233] 3. Path Selection: 3.1 Safe Distance Check: Calculate the minimum distance between each candidate path and the restricted area, ensuring all paths meet the requirement of a safe distance ≥ preset value. For example, if the preset safe distance is 5m, check whether each path maintains a distance of at least 5m from the restricted area. 3.2 Energy Consumption Estimation: Based on the path length and UAV performance parameters (such as flight speed, payload capacity, etc.), estimate the estimated energy consumption for each path. Ensure that the estimated energy consumption ≤ the preset percentage of remaining battery power. For example, if the remaining battery power is 60% and the preset percentage is 70%, then the estimated energy consumption must not exceed 70% of the remaining battery power.

[0234] 4. Path sorting:

[0235] Comprehensive Evaluation: A comprehensive evaluation is performed on the paths that meet the constraints, considering factors such as path length, flight time, and energy consumption. The optimal path is then selected after ranking. Optimal Path Output: The path with the highest comprehensive evaluation score is selected as the final alternative path and prepared for distribution to the drone.

[0236] In step S4E5, the generated candidate alternative paths are quickly simulated and verified using a lightweight scenario through a digital twin platform. The smoothness of the path, the risk of airspace conflict, and the timeliness of the mission are evaluated simultaneously. The alternative path with the best overall performance is selected and commands are issued to the UAV.

[0237] The specific process is as follows:

[0238] 1. Import candidate paths:

[0239] Path data: Import the candidate alternative paths generated in step S4E4 into the lightweight scenario of the digital twin platform. Path data includes waypoint coordinates, flight speed, and estimated time points.

[0240] Scene construction: Construct the drone's current position, target position, restricted area, and other relevant airspace elements in a lightweight scene.

[0241] 2. Rapid Simulation Verification: Path Smoothness Evaluation: Path smoothness is evaluated by calculating the curvature variation of the path. The smaller the curvature variation, the higher the smoothness. For example, the standard deviation of path curvature can be used as a smoothness index.

[0242] Among them, k i It is the curvature of the i-th point on the path. is the average curvature, and n is the number of path points.

[0243] Airspace conflict risk assessment: Airspace conflict risk is assessed by checking whether the path overlaps with the trajectories of known airspace users. For example, the minimum distance between the path and the known airspace user trajectories is calculated; if the minimum distance is less than a safety threshold, a conflict risk is considered to exist.

[0244] Mission timeliness achievement rate assessment: The mission timeliness achievement rate is assessed by calculating the ratio of the estimated flight time of the route to the remaining mission time.

[0245] 3. Comprehensive score: Calculate the comprehensive performance score of each candidate path based on path smoothness, airspace conflict risk and mission timeliness achievement rate.

[0246] For example, the overall performance score S can be expressed as: S = W1 × P + W2 × R + W3 × T.

[0247] Among them, w1, w2, and w3 are preset weights, representing the relative importance of path smoothness, airspace conflict risk, and task timeliness achievement rate, respectively.

[0248] 4. Select the optimal path:

[0249] Sorting: The candidate paths are sorted according to their overall performance scores, and the path with the lowest score is the optimal path.

[0250] Verification: Perform final verification on the optimal path to ensure that it meets all constraints, including safety distance and energy consumption requirements.

[0251] 5. Issue instructions:

[0252] Generate execution instructions: Convert the optimal path into a set of instructions that the UAV can execute, including waypoint coordinates, flight speed, and sensor operating modes.

[0253] Transmitting instructions: The instruction set is transmitted to the drone via the wireless communication module, guiding it to perform inspection tasks according to the new path.

[0254] Before establishing task-hive dynamic adaptation rules based on inspection task priorities and combining preset drone hive performance thresholds, the process also includes optimizing the spatial distribution of drone hives, as detailed below:

[0255] Step SA00: Extract the generated inspection task priority sequence, the spatial distribution of engineering entities in the digital twin benchmark model, and external environmental constraints.

[0256] The specific process is as follows:

[0257] 1. Extract the priority sequence of inspection tasks:

[0258] Retrieve from the task management system: Extract the generated inspection task priority sequence from the inspection task management system. This sequence is sorted according to the comprehensive priority score, reflecting the execution priority of each inspection task. For example, task A has the highest priority, followed by task B, and task C has the lowest priority.

[0259] 2. Extract the spatial distribution of engineering entities:

[0260] Obtain from the digital twin baseline model: Extract the spatial distribution data of engineering entities from the digital twin baseline model, including the geographical coordinates (such as latitude, longitude and altitude) of each entity. For example, the track signal equipment is located at (40.123456, 116.123456, 10m), and the substation is located at (40.123457, 116.123457, 5m).

[0261] 3. Extract external environmental constraints:

[0262] Data is obtained from external systems: External environmental constraint data is extracted from external systems such as meteorological service interfaces, air traffic control systems, and terrain databases. For example, meteorological conditions show that the wind speed in a certain area exceeds 10 m / s, the no-fly zone covers (40.123458, 116.123458, 10 m), and the terrain obstacle includes a hill with a height of 50 m.

[0263] Step SB00: Using a preset distribution rationality assessment model, calculate the average response distance of existing hives to high-priority task areas, the load rate variance of each hive, and the proportion of tasks in the coverage blind spots. When the response distance exceeds a preset threshold, the load rate variance is greater than a preset value, or the proportion of tasks in the blind spots is greater than or equal to a preset ratio, it is determined that the hive distribution needs to be optimized.

[0264] The distribution rationality assessment model is used to evaluate whether the current drone swarm distribution is reasonable. Average response distance: The average distance from existing swarms to high-priority task areas, reflecting the swarm's response speed to high-priority tasks. Load rate variance: The variance of the task load rate of each swarm, reflecting the balance of task allocation among swarms. Coverage blind spot task proportion: The proportion of high-priority task areas not covered by any swarms to the total task area. Preset threshold: A preset standard value used to determine whether swarm distribution needs to be optimized.

[0265] The specific process is as follows:

[0266] 1. Extract high-priority task regions:

[0267] Extract from the inspection task priority sequence: Obtain the area information of high-priority tasks, including geographical coordinates and task range. For example, the high-priority task areas include (40.123456, 116.123456, 10m) and (40.123457, 116.123457, 5m).

[0268] 2. Calculate the average response distance:

[0269] Extracting hive locations: Obtain the location information of existing hives from the digital twin baseline model. For example, hive 1 is located at (40.123458, 116.123458, 10m), and hive 2 is located at (40.123459, 116.123459, 10m).

[0270] Calculate distance: For each high-priority task area, calculate its distance to the nearest hive. For example, the distance from task area 1 to hive 1 is 0.5km, and the distance from task area 2 to hive 2 is 0.8km.

[0271] Calculate the average response distance: Take the average distance from all high-priority task areas to the nearest hive. For example, the average response distance is 0.65 km.

[0272] 3. Calculate the load factor variance:

[0273] Extract Hive Load Rate: Retrieve the current task load rate of each hive from the task management system. For example, the load rate of hive 1 is 70%, and the load rate of hive 2 is 30%.

[0274] Calculate the variance: Calculate the variance of the load factor for each hive. For example, the load factor variance is 0.16.

[0275] 4. Calculate the percentage of tasks with coverage blind spots: Identify coverage blind spots: Identify high-priority task areas from the digital twin baseline model that are not covered by any cellular data. For example, task area 3 is not covered by any cellular data. Calculate the percentage: Calculate the proportion of tasks with coverage blind spots out of the total number of high-priority tasks. For example, if there are 3 high-priority tasks in total, the percentage of tasks with coverage blind spots is 33.33%.

[0276] 5. Assess the rationality of the distribution:

[0277] Set a preset threshold: Preset average response distance threshold D threshold =0.5km, load factor variance threshold V threshold =0.1, threshold B for the percentage of tasks covering blind spots threshold =10%.

[0278] Determine if optimization is needed: If the average response distance D ≥ D threshold Or the load factor variance V ≥ V threshold Or the percentage of tasks covering blind spots is B≥B threshold If so, it is determined that the hive distribution needs to be optimized.

[0279] Step SC00 calls the preset site selection optimization algorithm, aiming to minimize the response distance of high-priority tasks, balance the load of the entire hive, and minimize the coverage blind spot. It combines the distribution of engineering entities, environmental constraints, and the flight radius of the UAV to generate an optimized distribution scheme that includes the coordinates of newly added hive points and the migration coordinates of existing hives.

[0280] In step SC00, optimization objectives are first set, including minimizing the average response distance from high-priority task areas to the nearest cellular unit, balancing the task load rate of each cellular unit, and minimizing the proportion of uncovered high-priority task areas. Specifically, the average response distance is measured by calculating the mean of the Euclidean distances from all high-priority task areas to the nearest cellular unit; load balancing is evaluated by calculating the standard deviation of the number of tasks for each cellular unit; and the coverage blind spot ratio is determined by calculating the proportion of high-priority task areas not covered by any cellular unit to the total task area. Next, input data is extracted, including the geographic coordinates of high-priority task areas, the spatial distribution of engineering entities, external environmental constraints (such as weather conditions, no-fly zones, and terrain obstacles), and key parameters such as the UAV's endurance radius.

[0281] Then, a preset site selection optimization algorithm is invoked, and the extracted data is input into the algorithm. During runtime, the algorithm generates multiple potential hive distribution schemes based on the optimization objectives. These schemes include not only suggested locations for new hives but also the migration coordinates of existing hives. In generating the schemes, the algorithm comprehensively considers the distribution of engineering entities to ensure hives are close to critical equipment; it incorporates environmental constraints to avoid placing hives in no-fly zones or areas with severe weather; and it considers the drone's endurance radius to ensure the feasibility and efficiency of the inspection mission. For example, the algorithm might use spatial analysis techniques to identify distribution hotspots in high-priority task areas and combine this with the drone's endurance to determine the optimal hive location.

[0282] Finally, each potential solution is comprehensively evaluated to check whether it meets the optimization objectives. For example, by simulating the allocation of inspection tasks, the average response distance, load rate variance, and coverage blind spot ratio of each solution are calculated. The optimal solution is selected, which should have the shortest average response distance, the smallest load rate variance, and the lowest coverage blind spot ratio.

[0283] Step SD00 involves inputting the optimized distribution scheme into the digital twin baseline model for simulation. This simulation assesses the overlap of the cellular coverage area, task allocation efficiency, and emergency scheduling feasibility under different task priorities, ultimately selecting the optimal distribution scheme.

[0284] In step SD00, the optimized distribution scheme is first input into the digital twin baseline model, which specifically includes the newly added hive locations and the migration coordinates of existing hives.

[0285] For example, the new hive location is (40.123459, 116.123459, 10m), and the existing hive 1 is moved to (40.123460, 116.123460, 10m). The overlap of the coverage areas of each hive is calculated by the model. The overlapping area of ​​the coverage areas can be calculated using the geometric intersection algorithm, resulting in an overlap of 0.2. When evaluating the efficiency of task allocation, the variance of the task load rate of each hive is calculated. Assuming a variance of 0.05, it indicates that the task allocation is relatively balanced. Sudden task scenarios are simulated to verify the rapid response capability of the hives. For example, when a sudden situation occurs in a high-priority task area, can the hives dispatch drones to the scene within a specified time?

[0286] Next, simulations were conducted to examine the overlap of cellular coverage, task allocation efficiency, and emergency dispatch feasibility under different task priorities. In the coverage overlap simulation, a geometric intersection algorithm was used to accurately calculate the overlap area of ​​each cellular's coverage region, and this area was compared with the total area to obtain an overlap index. The task allocation efficiency simulation, based on a load balancing algorithm, calculated the number of tasks assigned to each cellular and derived the load rate variance to assess balance. The emergency dispatch feasibility simulation simulated sudden task scenarios, considering the UAV's flight speed and endurance, to verify whether the cellular can respond and complete tasks within a specified time.

[0287] Finally, the comprehensive evaluation score for each optimized scheme is calculated based on the simulation results. The comprehensive evaluation score S can be expressed as: .

[0288] W1, W2, and W3 are preset weights, representing the relative importance of coverage overlap, task allocation efficiency, and emergency dispatch feasibility, respectively. All optimized schemes are ranked according to their comprehensive evaluation scores, and the scheme with the highest score is selected as the optimal distribution scheme, ensuring it performs best in terms of coverage, task allocation, and emergency dispatch. For example, scheme C has a comprehensive evaluation score of 0.90, higher than scheme A's 0.85 and scheme B's 0.75; therefore, scheme C is selected as the optimal scheme.

[0289] Step SE00: Complete the hive deployment adjustment according to the optimal solution, and synchronously update the preset drone hive performance threshold association information and hive spatial attributes in the digital twin benchmark model.

[0290] In addition, after preprocessing at the edge, the data is uploaded to the digital twin platform, where a pre-defined spatiotemporal data assimilation engine fuses it with the digital twin baseline model, driving dynamic model iteration. This specifically includes:

[0291] S1. By deploying a lightweight AI model at the edge of the drone, the system performs on-orbit real-time analysis of multi-dimensional real-time data obtained from inspections, accurately identifying equipment defects and abnormal states.

[0292] During drone inspections, the system utilizes lightweight AI models deployed at the drone's edge to perform real-time on-orbit analysis of acquired multi-dimensional data, accurately identifying equipment defects and abnormal states. These lightweight AI models, such as MobileNet or Tiny-YOLO, are optimized for efficient operation on resource-constrained edge devices and are trained using deep learning algorithms to identify various equipment defects and abnormal states, such as cracks, peeling, and leaks. The drone collects multi-dimensional data in real-time, including high-definition images, infrared thermal imaging, and laser scanning. The AI ​​model analyzes this data in real-time, extracting key features to accurately identify equipment defects and abnormal states. The identification results are then fed back to the drone control system in real-time, enabling immediate decision-making based on preset rules, such as adjusting flight paths or increasing shooting frequency, to ensure timely handling of defects and abnormal states.

[0293] S2. Based on the pre-built defect knowledge base and preset risk assessment rules, the identified equipment defects and abnormal states are encapsulated into standardized risk events, and risk levels are classified, and real-time alarm information is generated simultaneously.

[0294] Specifically, after identifying equipment defects and abnormal states, the system encapsulates these identification results into standardized risk events based on a pre-built defect knowledge base and preset risk assessment rules, and then classifies them into risk levels. The defect knowledge base records various defect types and their characteristic descriptions in detail, while the risk assessment rules classify defects into different risk levels, such as low risk, medium risk, and high risk, based on their specific characteristics and types. The system evaluates each identified defect according to these rules and generates corresponding real-time alarm information, including the alarm level, defect location, and a detailed description, so that maintenance personnel can respond quickly.

[0295] S3. The data stream, AI recognition results, and hierarchical alarm information that have undergone edge preprocessing are simultaneously uploaded to the digital twin platform via the communication network.

[0296] After edge preprocessing and AI recognition, the system integrates the processed data stream, AI recognition results, and tiered alarm information into structured data packets. These data packets are synchronously uploaded to the digital twin platform via high-speed, reliable communication networks, such as 5G or satellite communication. Data encryption and verification mechanisms are employed during communication to ensure the security and reliability of data transmission. The digital twin platform receives and parses these data packets, storing multi-dimensional real-time data, recognition results, and alarm information in a distributed database to support subsequent analysis and decision-making.

[0297] S4. Using a pre-set spatiotemporal data assimilation engine, the received multi-source heterogeneous data is deeply fused with the digital twin benchmark model to drive the model to update state parameters and correct geometric properties.

[0298] Specifically, after receiving multi-source heterogeneous data, the digital twin platform uses a pre-defined spatiotemporal data assimilation engine to deeply fuse this data with the digital twin baseline model. The spatiotemporal data assimilation engine processes multi-source heterogeneous data, including image data, sensor data, and alarm information, through spatiotemporal analysis algorithms, ensuring the consistency and accuracy of the data with the baseline model. During the fusion process, the model's state parameters, such as equipment health and defect location, as well as geometric attributes, such as crack length and width, are updated and corrected to reflect the current state of the equipment and changes in defects.

[0299] S5. Based on the updated high-fidelity digital twin benchmark model, the evolution of the risk events that have been connected is simulated, and the development trajectory and potential risk consequences under the influence of different external environmental factors are simulated through simulation algorithms that integrate physical mechanisms.

[0300] Based on the updated high-fidelity digital twin benchmark model, the system performs evolutionary simulations of the connected risk events. Through simulation algorithms integrating physical mechanisms, such as finite element analysis and fluid dynamics simulation, it simulates the development trajectory and potential risk consequences of risk events under the influence of different external environmental factors (such as temperature, humidity, and wind speed). These simulation algorithms consider the impact of physical laws and environmental factors on risk events, and can predict the expansion trend of defects and potential safety issues. Through evolutionary simulation, the system generates a risk evolution map, displaying the state changes and risk levels of risk events at different time points.

[0301] S6. Based on the risk evolution map and impact assessment results obtained from the deduction, reverse verification and automatic parameter calibration are performed on the equipment health calculation model, risk weight parameters and inspection strategies involved in the multi-dimensional assessment algorithm.

[0302] The technical process is as follows: 1. Risk Evolution Map Analysis: The system analyzes and extrapolates the risk evolution map to assess the development trend and potential risks of defects. For example, if the map shows that a crack in a piece of equipment is expanding, the system will record this change and prepare for subsequent verification and calibration. 2. Reverse Verification of Equipment Health Calculation Model: The system retrospectively verifies whether the equipment health calculation model accurately reflects the changes in crack expansion. For example, if the model fails to reflect crack expansion in a timely manner, the system will adjust the model parameters, such as increasing the weight of crack length and width, to more accurately assess the equipment health status. The specific formula is as follows: Equipment Health = w1 × Crack Length + w2 × Crack Width + w3 × Other Factors. Where w1, w2, and w3 are model parameters, adjusted according to the risk evolution map.

[0303] 3. Calibrate Risk Weight Parameters: The system calibrates risk weight parameters based on actual risk consequences. For example, if the crack propagation rate has a significant impact on equipment safety, the system will increase the weight corresponding to the crack propagation rate. The specific formula is as follows: Risk Score = w1 × Crack Length + w2 × Crack Width + w3 × Crack Propagation Rate + w4 × Other Factors. Where w1, w2, w3, and w4 are risk weight parameters, adjusted according to the risk evolution map and impact assessment results.

[0304] Optimize Inspection Strategy: Based on the results of reverse verification and parameter calibration, the system optimizes the inspection strategy. For example, if a crack in a certain area is spreading rapidly, the system will increase the inspection frequency for that area and adjust the drone's flight path to more efficiently cover critical areas. Specific measures include: 1. Increasing Inspection Frequency: For high-risk areas, the number of inspections is increased to ensure timely detection of new defects or changes. 2. Adjusting Flight Paths: The drone's flight path is optimized to enable it to more efficiently cover critical areas and reduce inspection time.

[0305] Automated calibration mechanism: The system employs an automated calibration mechanism that automatically adjusts parameters based on validation results. The calibrated parameters are then updated in the evaluation algorithm via a feedback mechanism, improving the algorithm's performance and adaptability. For example, the system can use machine learning algorithms, such as linear regression or neural networks, to automatically adjust parameters, ensuring the model's accuracy and reliability.

[0306] S7. Through the iterative process described above, a closed-loop intelligent evolution mechanism is formed, from on-site perception and accurate diagnosis to decision optimization, thereby continuously improving the self-evolution capability of the digital twin system.

[0307] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A digital twin control method for rail transit engineering based on UAV cellular inspection, characterized in that, include: Acquire multi-source data for rail transit engineering, including BIM design models, GIS geographic information, IoT sensor data, and initial UAV imagery; Extract the core parameters of the BIM design model, perform lightweight processing based on these parameters, and integrate multi-scale LOD rendering technology to adapt to scene requirements. Complete the registration with GIS geographic information through a preset coordinate system fusion algorithm, and establish a dynamic association network of engineering entities, environment and events by combining spatiotemporal tagging technology to form a digital twin benchmark model. Based on real-time status data from a digital twin benchmark model, external factors are integrated, and a pre-set multi-dimensional evaluation algorithm is used to generate inspection task priorities. The real-time status data includes equipment health values ​​and environmental monitoring data, while the external factors include real-time weather, temporary construction areas, and train running times. Based on the preset performance thresholds of the drone hive, a task-hive dynamic adaptation rule is established according to the priority of the inspection task. Drones are scheduled to perform inspections by avoiding restricted areas according to the preset path optimization algorithm. The performance thresholds include wind resistance level, endurance margin, and load capacity. The above-mentioned inspections obtain multi-dimensional real-time data, which is then preprocessed at the edge and uploaded to the digital twin platform. The pre-set spatiotemporal data assimilation engine merges the data with the digital twin benchmark model, driving the model to iterate dynamically and achieve real-time synchronous updates between the twin and the engineering site.

2. The digital twin control method for rail transit engineering based on UAV cellular inspection as described in claim 1, characterized in that, Based on real-time status data from a digital twin benchmark model, and incorporating external factors, a pre-defined multi-dimensional evaluation algorithm is used to generate inspection task priorities, including: Extract real-time status data from the digital twin baseline model and synchronously correlate it with equipment importance level, historical failure frequency, and external factors; The real-time status data, equipment importance level, historical failure frequency and external factors are normalized. The difference in units is eliminated by a preset standardization algorithm, and the data that deviates from the reasonable range is removed by a preset outlier filtering algorithm to form a standardized dataset. Based on a standardized dataset, a preset dynamic weighting algorithm is used to allocate weights for each dimension. The equipment importance level is coupled with the equipment health value in the real-time status data as the core weight. The weight of historical failure frequency is adjusted according to the time decay coefficient, and external factors are given correction weights according to their degree of influence. The weighted data of each dimension are input into a preset spatiotemporal correlation model. The fault transmission correlation degree of adjacent equipment is calculated by combining the spatial distribution characteristics of engineering entities, and an evaluation matrix containing spatiotemporal influence factors is generated. By integrating the feature values ​​of the evaluation matrix data and the standardized dataset through a preset fusion algorithm, the comprehensive priority score of each potential inspection task is calculated. The inspection tasks are sorted from highest to lowest based on their overall priority scores to generate a priority sequence.

3. The digital twin control method for rail transit engineering based on UAV cellular inspection as described in claim 2, characterized in that, Generating an evaluation matrix that includes spatiotemporal influence factors includes: The weighted data of each dimension are input into a preset spatiotemporal correlation model. This model is based on the spatial distribution characteristics of engineering entities and constructs a spatial topology network of engineering entities through a built-in spatial topology algorithm, identifies adjacent devices in the network and marks the physical connection type. The spatiotemporal correlation model is based on the labeled physical connection type, calls the built-in time correlation algorithm, introduces the preset time decay rule, and combines the weighted historical fault frequency data and historical fault propagation records to calculate the probability of fault propagation from the source device to the adjacent device in different time periods. Among them, the physical connection type directly affects the baseline value of the propagation probability. The spatiotemporal correlation model combines the identified physical connection types and the calculated transmission probabilities, and through the built-in correlation fusion algorithm, it fuses the spatial connection strength and temporal transmission characteristics to obtain the fault transmission correlation degree of adjacent devices. The spatiotemporal correlation model uses each device in the spatial topology network as the matrix row / column index, and the calculated fault propagation correlation degree as the matrix base value. It also superimposes a preset time decay factor and a preset device importance correction coefficient to generate an evaluation matrix that includes spatiotemporal influence factors.

4. The digital twin control method for rail transit engineering based on UAV cellular inspection according to claim 3, characterized in that, The calculation of the overall priority score for each potential inspection task includes: The feature values ​​of the assessment matrix data containing spatiotemporal impact factors and the standardized dataset are extracted. The assessment matrix data characterizes the intensity of spatiotemporal correlation risk between devices. Based on the characteristics of rail transit engineering, a preset fusion dimension weight allocation rule is adopted to assign preset weights to the equipment health and importance coupling value, historical fault and external factor correction value in the evaluation matrix data and the feature values ​​of the standardized dataset. Based on the extracted evaluation matrix data and the feature values ​​of the standardized dataset, a pre-set conflict resolution algorithm is used to handle potential contradictions between data, ensuring the consistency of data from different dimensions during fusion. The evaluation matrix data after conflict resolution is mapped to a single device dimension according to the device correspondence, and the basic score of each dimension is obtained by multiplying the feature value of the standardized dataset by the corresponding preset fusion dimension weight. For high-frequency risk areas and areas affected by external emergency factors that are centrally labeled in the standardized dataset, a preset dynamic adjustment factor is introduced to adjust the scores of tasks in high-frequency risk areas and tasks in areas affected by external emergency factors according to the preset coefficients. The comprehensive priority score for each potential inspection task is obtained by weighting and summing the base scores of each dimension with the corresponding dynamic adjustment factors.

5. The digital twin control method for rail transit engineering based on UAV cellular inspection according to claim 1, characterized in that, Based on the preset performance thresholds of the drone cellular network, and according to the priority of inspection tasks, dynamic adaptation rules for task-cellular networks are established, including: Extract the preset performance threshold parameters of the drone hive, the real-time status data of the drone, and the key parameters of the inspection task. The key parameters of the inspection task include the inspection area, the type of target equipment and the required detection accuracy, and the timeliness requirements corresponding to the task priority. Simultaneously acquire external factor data as constraints for adaptation rules. External factor data includes wind speed and precipitation levels from real-time meteorological data, real-time boundary update data of temporary construction areas, and no-fly periods information from train operation plans. Based on the inspection task priority, combined with the aforementioned performance threshold parameters, key task parameters and external factor data, a dynamic adaptation rule is established through a preset priority-performance matching logic. High-priority tasks are matched with hives that meet the core performance standards, while low-priority tasks are matched according to the principle of resource balancing. Unexpected factors trigger the switching of backup hives for high-priority tasks.

6. The digital twin control method for rail transit engineering based on UAV cellular inspection according to claim 5, characterized in that, Dispatching drones to perform inspections by using a preset path optimization algorithm to avoid restricted areas includes: Based on the matching results generated by the task-honeycomb dynamic adaptation rules, the spatiotemporal constraints of the restricted area are obtained from the digital twin benchmark model and external systems, including the coordinates and time periods of the train no-fly zone, the isolation boundary of the temporary construction area, and the range of the meteorological high-risk area, which are then transformed into the preset spatial and temporal constraints for path planning. Extract the performance parameters and inspection task requirements of the matching UAV and input them into the preset path optimization algorithm. The preset path optimization algorithm prioritizes meeting the time requirements and minimizing energy consumption. Combined with the aforementioned constraints and parameters, it generates an initial path that avoids all restricted areas, including the coordinates of the waypoints, flight speed and estimated time nodes. The initial path is simulated for compliance using a digital twin 3D scene. If there is overlap with the restricted area or insufficient safety distance, the algorithm is triggered to iterate and optimize again until the path fully complies with the constraints. Send execution instructions containing optimized paths and sensor operating modes to the matched drones, and schedule them to start inspections according to the paths; The system receives real-time updates on the drone's location, remaining battery power, and external factors. If a conflict is detected between the path and the updated restricted area, the system immediately invokes a preset dynamic obstacle avoidance sub-algorithm to generate an alternative path.

7. The digital twin control method for rail transit engineering based on UAV cellular inspection according to claim 6, characterized in that, If a conflict is detected between the path and the updated restricted area, the preset dynamic obstacle avoidance sub-algorithm is immediately invoked to generate an alternative path, including: Real-time reception of drone location data and remaining battery information, synchronous acquisition of dynamically updated data on external factors, forming the basis data for conflict analysis; Based on conflict analysis data, the spatial overlap and temporal conflict coefficient between the current planned path and the updated restricted area are calculated using a preset conflict determination model. When any indicator reaches the preset conflict threshold, the obstacle avoidance response mechanism is triggered. After the obstacle avoidance response mechanism is triggered, the core parameters in the basic data of conflict analysis are extracted and associated with the remaining time requirement of the task and the meteorological energy consumption correction parameters, which are used as the input data for the preset dynamic obstacle avoidance sub-algorithm. The preset dynamic obstacle avoidance sub-algorithm is invoked, with the core objectives of prioritizing obstacle avoidance safety and minimizing time loss. A preset number of candidate alternative paths are generated in combination with the input parameters. All paths must meet the dual constraints of a safe distance from the restricted area ≥ a preset value and an estimated energy consumption ≤ a preset proportion of the remaining power. For the generated candidate alternative paths, a lightweight scenario is quickly simulated and verified using a digital twin platform. The smoothness of the path, the risk of airspace conflict, and the timeliness of the mission are evaluated simultaneously. The alternative path with the best overall performance is selected and commands are issued to the UAV.

8. The digital twin control method for rail transit engineering based on UAV cellular inspection according to claim 1, characterized in that, Before establishing task-hive dynamic adaptation rules based on inspection task priorities and combining preset drone hive performance thresholds, the process also includes optimizing the spatial distribution of drone hives, as detailed below: Extract the generated inspection task priority sequence, the spatial distribution of engineering entities in the digital twin benchmark model, and external environmental constraints; Using a pre-defined distribution rationality assessment model, the average response distance of existing hives to high-priority task areas, the load rate variance of each hive, and the proportion of tasks in the coverage blind spots are calculated. When the response distance exceeds a pre-defined threshold, the load rate variance is greater than a pre-defined value, or the proportion of tasks in the blind spots is greater than or equal to a pre-defined proportion, it is determined that the hive distribution needs to be optimized. The preset site selection optimization algorithm is invoked to generate an optimized distribution scheme that includes newly added hive locations and the migration coordinates of existing hives, with the goals of minimizing the response distance of high-priority tasks, achieving full hive load balancing, and minimizing coverage blind spots, in combination with the distribution of engineering entities, environmental constraints, and the flight radius of UAVs. The optimized distribution scheme is input into the digital twin benchmark model for simulation, which simulates the overlap of the cellular coverage, task allocation efficiency and emergency scheduling feasibility under different task priorities, and selects the optimal distribution scheme. Complete the hive deployment adjustment according to the optimal solution, and simultaneously update the performance threshold association information of the preset drone hive and the hive spatial attributes in the digital twin benchmark model.

Images