Early warning method and system for load unbalance fault of electric power emergency multi-machine lifting in mountainous area

Abstract

The invention provides a mountainous area electric power emergency multi-machine lifting load unbalance fault early warning method and system, and relates to the technical field of data processing, the method comprises the steps that multi-source sensing data is mapped and converted into a three-dimensional space data array, and each data sampling position corresponds to a space position coordinate; on the basis of the three-dimensional space data array, a continuous balance curved surface representing the dynamic balance state of multi-machine lifting is constructed; determining vertical projection points of four unmanned aerial vehicle gravity center positions on the continuous balance curved surface on the continuous balance curved surface; constructing a four-boundary feature configuration region based on the vertical projection points; based on the four-boundary feature configuration region and the continuous equilibrium curved surface feature boundary intersection point, a geometric imbalance correction coefficient is solved; and generating a comprehensive load unbalance feature vector according to the geometric unbalance correction coefficient, the multi-source sensing data and the type characteristics of the currently hoisted power repair load. According to the invention, the safety and reliability of multi-machine cooperative hoisting operation in a complex mountainous area environment can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for early warning of load imbalance faults in multi-machine hoisting in mountainous power emergency situations. Background Technology

[0002] In emergency repairs of power lines in mountainous areas, using multiple drones to collaboratively lift power equipment or repair materials is an efficient and flexible method of material transportation. This multi-drone lifting system, where the load is connected by tethers, is crucial for dynamic stability and safety. However, the complex and changeable mountainous environment, with its gusts of wind and air currents, can easily cause the load to sway, leading to uneven stress on the drones, flight instability, and even load imbalance in the entire lifting system. If this imbalance worsens, it can cause tethers to break, the load to fall, or drones to collide, seriously threatening operational safety and repair efficiency.

[0003] Currently, existing technical solutions for status monitoring and early warning of such multi-machine hoisting systems mostly focus on threshold monitoring of single physical quantities (such as the tension of a single tether rope or the load tilt angle), or on independently judging the attitude of each UAV. These methods have some shortcomings: For example, these methods fragment sensor data from different sources and locations (drones, loads, environment), ignoring the inherent spatial correlation and synergistic influence mechanisms of multi-source data, and failing to grasp the overall equilibrium state of the system. Secondly, some existing methods lack consideration of the coupling relationship between the physical characteristics of the load itself (such as mass distribution and aerodynamic characteristics) and the dynamic imbalance process, failing to intuitively construct and evaluate the dynamic equilibrium status of the entire hoisting system on a three-dimensional spatial scale, thus limiting the accuracy of early warning. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for early warning of load imbalance faults in multi-machine hoisting in mountainous power emergency situations, which can improve the safety and reliability of multi-machine collaborative hoisting operations in complex mountainous environments.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas, the method comprising: Collect multi-source sensor data during multi-machine hoisting. The multi-source sensor data includes tether tension data of each UAV, load swing status data, flight attitude data of each UAV, and mountainous environmental disturbance data. Multi-source sensor data is mapped and converted into a three-dimensional spatial data array, with each data sampling location corresponding to a spatial coordinate. Based on a three-dimensional spatial data array, a continuous equilibrium surface is constructed to characterize the dynamic equilibrium state of multiple cranes. Determine the vertical projection points of the four UAV center of gravity positions onto the continuous equilibrium surface; Construct a four-boundary feature configuration region based on the vertical projection points; solve for the geometric imbalance correction coefficient based on the intersection of the four-boundary feature configuration region and the feature boundary of the continuous equilibrium surface. Based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the current power repair load type being hoisted, a comprehensive load imbalance feature vector is generated. Based on the comprehensive load imbalance feature vector, collaborative abnormal pattern recognition is performed on the comprehensive load imbalance feature vector to identify early trend characteristics of load imbalance. When the load imbalance risk is detected to reach the preset warning level, a load imbalance fault warning signal is generated in real time.

[0006] Secondly, a fault early warning system for multi-machine hoisting load imbalance in mountainous power emergency systems includes: The data acquisition module is used to collect multi-source sensor data during multi-machine hoisting. The multi-source sensor data includes the tether tension data of each UAV, the load swing status data, the flight attitude data of each UAV, and the environmental disturbance data in the mountainous area. The conversion module is used to map and convert multi-source sensor data into a three-dimensional spatial data array, with each data sampling position corresponding to a spatial coordinate. The module is used to construct a continuous equilibrium surface representing the dynamic equilibrium state of multiple cranes based on a three-dimensional spatial data array; and to determine the vertical projection points of the center of gravity of four UAVs on the continuous equilibrium surface. The solver module is used to construct a four-boundary feature configuration region based on vertical projection points; and to solve for the geometric imbalance correction coefficient based on the intersection of the four-boundary feature configuration region and the feature boundary of the continuous equilibrium surface. The processing module is used to generate a comprehensive load imbalance feature vector based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the current power repair load type being hoisted. The early warning module is used to perform collaborative abnormal pattern recognition on the comprehensive load imbalance feature vector based on the comprehensive load imbalance feature vector, identify early trend features of load imbalance, and generate a load imbalance fault early warning signal in real time when the load imbalance risk is detected to reach the preset early warning level.

[0007] The above-described solution of the present invention has at least the following beneficial effects: By constructing a continuous equilibrium surface and extracting spatial geometric features such as the four-boundary feature configuration region and feature boundary intersections, this invention can map the microscopic mechanical and kinematic changes of the system into subtle changes in the macroscopic surface morphology and geometric relationships. The solved geometric imbalance correction coefficient is a sensitive quantitative indicator of the deviation from the spatial equilibrium state. This method has a strong ability to capture early and weak signals of system imbalance, realizing the transformation from ex-post alarm based on obvious thresholds to early trend warning based on spatial morphological evolution, thus gaining valuable time for proactive regulation.

[0008] The comprehensive load imbalance feature vector proposed in this invention not only integrates geometric correction coefficients characterizing macroscopic spatial imbalance states, but also coordinates time-frequency domain dynamic features from multi-source sensors, and further incorporates the type and characteristic parameters of the current hoisting load. This multi-level, multi-dimensional feature fusion approach makes the description of the system state more comprehensive and accurate. The subsequent collaborative anomaly pattern recognition does not simply judge whether a single feature exceeds its limit, but rather deeply analyzes the coupling relationships and collaborative evolution trends between multiple features. This effectively distinguishes between normal fluctuations caused by environmental disturbances and anomaly patterns characterizing true instability risks, significantly reducing the false alarm rate. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a method for early warning of load imbalance faults in multi-machine hoisting in mountainous power emergency, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of a multi-machine hoisting load imbalance fault early warning system for mountain power emergency provided by an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] like Figure 1 As shown in the figure, an embodiment of the present invention proposes a method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas. The method includes the following steps: Step 1: Collect multi-source sensor data during multi-machine hoisting. The multi-source sensor data includes the tether tension data of each UAV, load swing status data, flight attitude data of each UAV, and mountainous environmental disturbance data. Step 2: Map the multi-source sensor data into a three-dimensional spatial data array, with each data sampling location corresponding to a spatial coordinate. Step 3: Based on the three-dimensional spatial data array, construct a continuous equilibrium surface that characterizes the dynamic equilibrium state of multiple cranes; Step 4: Determine the vertical projection points of the four UAV center of gravity positions on the continuous equilibrium surface; Step 5: Construct a four-boundary feature configuration region based on the vertical projection points; solve for the geometric imbalance correction coefficient based on the intersection of the four-boundary feature configuration region and the feature boundary of the continuous equilibrium surface. Step 6: Generate a comprehensive load imbalance feature vector based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the current power emergency repair load type being hoisted. Step 7: Based on the comprehensive load imbalance feature vector, perform collaborative abnormal pattern recognition on the comprehensive load imbalance feature vector to identify early trend features of load imbalance. When the load imbalance risk is detected to reach the preset warning level, generate a load imbalance fault warning signal in real time.

[0012] In this embodiment, the present invention innovatively maps and fuses heterogeneous, multi-source sensor data from UAVs, loads, and the environment into a three-dimensional spatial data array through a unified global spatial reference system. This processing method breaks the limitation of data fragmentation in traditional methods, so that all key parameters such as tension, attitude, swing, and disturbance are given clear spatial attributes and positional associations. Thus, the balance status of the entire multi-machine hoisting system can be dynamically and intuitively constructed and characterized from the perspective of the overall spatial relationship.

[0013] By constructing a continuous equilibrium surface and extracting spatial geometric features such as the four-boundary feature configuration region and feature boundary intersections, this invention can map the microscopic mechanical and kinematic changes of the system into subtle changes in the macroscopic surface morphology and geometric relationships. The solved geometric imbalance correction coefficient is a sensitive quantitative indicator of the deviation from the spatial equilibrium state. This method has a strong ability to capture early and weak signals of system imbalance, realizing the transformation from ex-post alarm based on obvious thresholds to early trend warning based on spatial morphological evolution, thus gaining valuable time for proactive regulation.

[0014] The comprehensive load imbalance feature vector proposed in this invention not only integrates geometric correction coefficients characterizing macroscopic spatial imbalance states, but also coordinates time-frequency domain dynamic features from multi-source sensors, and further incorporates the type and characteristic parameters of the current hoisting load. This multi-level, multi-dimensional feature fusion approach makes the description of the system state more comprehensive and accurate. The subsequent collaborative anomaly pattern recognition does not simply judge whether a single feature exceeds its limit, but rather deeply analyzes the coupling relationships and collaborative evolution trends between multiple features. This effectively distinguishes between normal fluctuations caused by environmental disturbances and anomaly patterns characterizing true instability risks, significantly reducing the false alarm rate.

[0015] In a preferred embodiment of the present invention, step 1 involves collecting multi-source sensor data during multi-machine hoisting. This multi-source sensor data includes tether tension data for each UAV, load sway status data, flight attitude data for each UAV, and environmental disturbance data in mountainous areas. High-precision tension sensors are installed at the attachment points and tether connections of each UAV. These sensors measure the tension on each tether in real time and transmit the tension data to the central processing unit via the UAV's onboard data link. The collected data includes the instantaneous tension value, timestamp, and corresponding UAV number. An attitude reference system with an integrated inertial measurement unit is installed at the load's center of mass or key structural points. This system incorporates a gyroscope and accelerometer to measure the load's angular velocity, angular acceleration, pitch angle, roll angle, and yaw angle in three-dimensional space in real time. This data directly characterizes the load's swing amplitude, frequency, and direction, and is wirelessly transmitted to the central processing unit via a data transmission module onboard the load. Each UAV's own flight control system continuously outputs high-precision flight attitude data, including at least the UAV's pitch angle, roll angle, yaw angle, three-axis angular velocity, and three-axis acceleration. This data, along with the tether tension data, is packaged and transmitted to the central processing unit via the UAV's onboard data link to characterize the spatial attitude stability of each UAV.

[0016] Miniature weather stations or ultrasonic anemometers are deployed at key locations within the operational area (e.g., the navigation drone, the payload itself, or a ground base station). These devices monitor environmental disturbance parameters in real time, such as wind speed, instantaneous wind direction, gust intensity, and airflow change rate. Environmental data is synchronously transmitted to the central processing unit via a wireless network. Upon receiving sensor data streams from all sources, the central processing unit first timestamps all data according to a high-precision, unified clock source. Subsequently, following a pre-defined data frame structure, it aligns, packages, and integrates the tether tension data, flight attitude data, payload sway status data, and mountainous environmental disturbance data collected simultaneously from each drone, forming a structured, time-stamped multi-source sensor data packet.

[0017] In a preferred embodiment of the present invention, step 2, mapping and converting multi-source sensor data into a three-dimensional spatial data array, wherein each data sampling position corresponds to a spatial position coordinate, includes: Step 21: Based on the load swing state data, determine the instantaneous center of gravity position of the load, and construct a global spatial reference system using the instantaneous center of gravity position as the origin. Within the global spatial reference system, calculate the spatial position coordinates of each UAV, specifically including: determining the instantaneous center of gravity position of the load based on the load swing state data; the load attitude reference system outputs real-time data such as angular velocity, angular acceleration, pitch angle, roll angle, and yaw angle in three-dimensional space, combined with the load's preset physical parameters (including the overall mass of the load, known geometric dimensions, and pre-calibrated mass distribution data of each component), and calculates the instantaneous center of gravity coordinates of the load using a center of gravity calculation algorithm. The algorithm uses the design reference center point of the load as the initial reference. Based on the changes in the load's spatial attitude reflected by the attitude data, it dynamically corrects the relative positions of each component, and finally obtains the instantaneous three-dimensional coordinates (X0, Y0, Z0) of the load's center of gravity in the geodetic coordinate system. Then, a global spatial reference system is constructed with this instantaneous center of gravity position as the origin (0, 0, 0). The X-axis points to the preset forward direction of the hoisting operation, the Y-axis is perpendicular to the X-axis and lies in the horizontal plane, and the Z-axis is perpendicular to the XY plane to form a right-handed coordinate system. This coordinate system is updated in real time with the dynamic changes of the load's center of gravity, ensuring that the load is always the spatial core for establishing the reference reference.

[0018] When calculating the spatial position coordinates of each UAV in the global spatial reference frame, it is necessary to combine multi-source data for fusion calculation. First, obtain the geodetic coordinates output by each UAV through the onboard GPS / BeiDou positioning module. , , Simultaneously, combining the pitch, roll, and yaw angles output by the UAV flight control system, as well as the tether tension state indirectly reflected by the tether tension sensor (excluding the influence of tether elastic deformation on distance calculation), the coordinates of the UAV in the geodetic coordinate system are transformed to a global spatial reference system with the instantaneous center of gravity of the load as the origin through a coordinate transformation algorithm, so as to obtain the real-time spatial position coordinates of each UAV in this reference system. , , ).

[0019] Step 22: Based on the spatial position coordinates of each UAV, calculate the mapped coordinates of the connection point of each tether above the load in the global spatial reference system; associate the corresponding UAV tether tension data with these mapped coordinates to obtain the first data; associate the load swing state data with the origin coordinates of the global spatial reference system to obtain the second data; associate the flight attitude data of each UAV with its own spatial position coordinates to obtain the third data; based on the actual measurement location and direction of the mountainous environmental disturbance data, determine the corresponding spatial influence source coordinates in the global spatial reference system, and associate the environmental disturbance data with the spatial influence source coordinates to obtain the fourth data, specifically including: Based on the spatial coordinates of each UAV ( , , Combining the structural design parameters of the drone's mounting points (including preset data such as the offset of the mounting point relative to the drone's center of gravity and the installation angle), geometric calculations are used to obtain the mapped coordinates (X, X) of each tether's connection point above the load (i.e., the tether mounting point on the load) in the global space reference system. t Y t Z t This calculation needs to consider the natural drooping characteristics of the tether in a static state and dynamically correct the spatial position of the attachment point by combining load attitude data. Subsequently, the instantaneous tension value, timestamp, and drone number collected by the corresponding drone tether tension sensor are associated one by one with the mapped coordinates of the connection point to form the first data containing spatial coordinates and tension parameters.

[0020] Since the instantaneous center of gravity of the load is the origin (0, 0, 0) of the global spatial reference system, and the load swing state data (angular velocity, angular acceleration, pitch angle, etc.) directly reflect the motion state of the load's center of gravity and the whole, all swing state data output by the load attitude reference system are directly associated with the origin coordinate position of the global spatial reference system to form second data with the origin as the spatial identifier. The content of the second data includes swing parameters, acquisition timestamps, and data validity identifiers.

[0021] The flight attitude data of each UAV (pitch angle, roll angle, yaw angle, three-axis angular velocity, and three-axis acceleration, etc.) directly reflects its own spatial motion state and is strongly correlated with the UAV's real-time spatial position. Therefore, the attitude data output by each UAV's flight control system, along with its own position coordinates in the global spatial reference frame, are used to determine its position. , , The data is bound together to ensure that each set of attitude parameters can be accurately mapped to the drone at a specific spatial location, forming the third data of drone coordinates-attitude parameters, along with the drone number and data synchronization timestamp.

[0022] Based on the actual measurement locations of environmental disturbance data in mountainous areas, the spatial coordinates of the sources of influence are determined for different scenarios. If the measurement equipment is deployed on a navigation drone, the coordinates of the navigation drone's position in the global spatial reference system (X...) are used. a ', Y a ', Z aThe coordinates of the source of environmental disturbance are used as follows: if deployed on a load, the coordinates of the origin of the global spatial reference system are used as the source coordinates; if deployed on a ground base station, the geodetic coordinates of the ground base station are first obtained, and then converted with the geodetic coordinates of the instantaneous center of gravity of the load to obtain its coordinates (X9', Y9', Z9') in the global spatial reference system as the source coordinates. At the same time, combined with the wind direction parameter in the environmental data, the propagation direction vector of the disturbance is marked on the source coordinates. The disturbance parameters such as wind speed, gust intensity, and airflow change rate are associated with the source coordinates and direction vector to form the fourth data of source coordinates-disturbance parameters-direction vector, ensuring the spatial attribute integrity of the environmental disturbance data.

[0023] Step 23 involves aggregating the first, second, third, and fourth data sets to obtain a three-dimensional spatial data slice. Specifically, this includes calibrating the timestamps contained in the first, second, third, and fourth data sets using a preset unified clock source as a reference, ensuring that the time base of all data is completely consistent. Subsequently, according to a preset sampling period (set based on the dynamic characteristics of the hoisting operation and the sensor response speed, typically 50ms-200ms), the calibrated multi-source data is divided into time-dimensional slices, extracting all data at the same sampling moment.

[0024] For data extracted at the same time, a spatial grid index table is established using the coordinates of the global spatial reference system as an index. This index table divides the global space with a fixed spatial precision (set according to the operation range, usually a 0.1m×0.1m×0.1m grid). Each grid corresponds to a unique index value, mapping the spatial coordinates (connection point coordinates, origin coordinates, UAV coordinates, and influence source coordinates) in each data unit to the corresponding spatial grid. The association between the data and the spatial grid is realized through the index value.

[0025] Next, the associated data within each spatial grid is integrated by attribute. The same grid may contain multiple types of data (e.g., the grid containing the coordinates of a certain UAV simultaneously contains the attitude data of the UAV and the tension data of the adjacent tether). These data are packaged according to the structure of coordinate information-data type-parameter value-validity identifier. The validity identifier is generated by a data verification algorithm and is used to determine whether there are problems such as transmission errors or sensor failures (e.g., tension data that exceeds the sensor's range is marked as invalid).

[0026] Finally, all spatial grids containing valid data are sorted by index value to form a structured data set containing all valid spatial correlation data at that moment, with spatial grids as the basic unit. This is a three-dimensional spatial data slice. The three-dimensional spatial data slice also contains metadata such as the sampling time timestamp corresponding to the slice and data integrity statistics (number of valid data units / total number of data units).

[0027] Step 24: Combine the 3D spatial data slices from multiple consecutive sampling times in chronological order to obtain a 3D spatial data array. Specifically, this involves: using the unified clock source used for data timestamp calibration in Step 23 as a reference, extracting the sampling timestamps contained in all 3D spatial data slices, treating each timestamp as an independent time node, and constructing a continuous global time axis in chronological order. The starting node of the time axis is the timestamp corresponding to the first sampled data slice, and the ending node is the timestamp corresponding to the last sampled data slice. The interval between two adjacent time nodes strictly matches the sampling period preset in Step 23 to ensure the uniformity and continuity of the time axis scale. For each 3D spatial data slice, through precise matching of its timestamp and time axis nodes, the slice is positioned at a unique corresponding position on the time axis, forming a one-to-one correspondence between time nodes and data slices, laying the temporal foundation for subsequent array combination. If a time node lacks a corresponding data slice due to temporary sensor failure or data transmission delay, the node is marked as missing data, and the specific time information of the missing moment is recorded, preserving a basis for subsequent data completion or anomaly analysis.

[0028] Secondly, a multi-dimensional structure of the three-dimensional spatial data array is constructed and data mapping is completed. Combining the spatial characteristics of hoisting operations and data association requirements, a three-dimensional structure system of time-space-attribute for the three-dimensional spatial data array is determined: The first dimension is the time dimension, directly corresponding to the constructed global time axis, with each dimension unit corresponding to a time node and a three-dimensional spatial data slice under that node; the second dimension is the spatial dimension, based on the spatial grid index table established in step 23, using all spatial grids in the global space divided with a preset precision as units of this dimension, with each unit corresponding to a unique spatial grid index value, ensuring complete alignment between the spatial dimension and the spatial attributes in the data slices; the third dimension is the attribute dimension, integrating various parameter attributes contained in the data slices in step 23, including tether tension parameters, load swing parameters, UAV attitude parameters, environmental disturbance parameters, and related information such as data validity identifiers and UAV numbers, with each attribute unit corresponding to a specific type of data parameter. After the structure definition is completed, the various types of data in the data slice at each time node are mapped one by one to the corresponding cells of the three-dimensional array according to the logic of time node positioning - spatial grid matching - attribute parameter correspondence. For example, at a certain sampling time, the pitch angle data of UAV No. 1 in the spatial grid with index value 5 will be accurately mapped to the intersection of the time cell at that sampling time - the spatial cell with index value 5 - the UAV attitude parameter attribute cell, thus realizing the structured storage of data.

[0029] Next, the correlation of array data is strengthened and redundancy is handled. Since the data in the UAV hoisting system has inherent correlation characteristics, after data mapping, the data correlation within the array needs to be strengthened based on preset correlation rules. Specifically, this includes: establishing internal correlation identifiers for the position coordinates, attitude data, and corresponding tether tension data of the same UAV using the UAV number, ensuring that at any given time, all relevant data for a particular UAV can be quickly extracted using these identifiers; establishing spatial correlation identifiers for the coordinates of the source of environmental disturbance data with the UAV and load data within that coordinate range, clarifying the impact range of environmental factors on equipment within a specific spatial area; and establishing a fixed correlation between load swing data and the origin coordinates of the global spatial reference system, ensuring that the load data always serves as the core spatial reference of the array. Simultaneously, for potentially redundant data in the array (such as timestamps and UAV numbers recorded repeatedly in different data types), a core information retention-redundancy identifier correlation approach is adopted. That is, only one copy of the core information is retained in the array's basic metadata, and the remaining correlated data points to the core information through redundancy identifiers, reducing data storage redundancy without compromising data correlation.

[0030] Finally, the integrity of the array and metadata supplementation are performed. A comprehensive integrity check is conducted on the combined 3D spatial data array. The check includes: whether the data slices of each node in the time dimension completely cover all sampling times from start to end, and whether the number and specific location of nodes with missing data are clearly recorded; whether the index values ​​of each spatial grid in the spatial dimension are unique and continuous, and whether there are any grid cells with incorrect data mapping; and whether various data parameters in the attribute dimension are complete, and whether there are any missing attributes or abnormal parameter values. For problems found during the check, such as data mapping errors, corrections are made; if attributes are missing, they are marked as incomplete attributes and associated with the corresponding error reason. After the check passes, complete metadata information is supplemented for the 3D spatial data array. In addition to the sampling times and data integrity statistics already included in each slice in step 23, new information such as the array's time span (the time difference between the start and end sampling times), spatial coverage (3D spatial boundary coordinates calculated based on the spatial grid index), total data volume, description of association identification rules, and sampling period are added, forming a complete array metadata archive. The final output three-dimensional spatial data array must contain both structured core data and complete metadata to ensure that subsequent applications such as dynamic stability analysis and control strategy optimization can quickly extract target data and clearly understand the source, attributes, and relationships of the data.

[0031] In a preferred embodiment of the present invention, step 3, constructing a continuous equilibrium surface characterizing the motional equilibrium state of the multi-crane crane based on a three-dimensional spatial data array, includes: Step 31: Extract the three-dimensional spatial data slice corresponding to the current sampling time from the three-dimensional spatial data array, obtain the spatial coordinates of all data sampling points in the three-dimensional spatial data slice, and mark the sensor data value associated with each coordinate point as the equilibrium state observation value of that point, thereby obtaining a set of labeled spatial points, specifically including: Based on the metadata information of the 3D spatial data array, the global timeline record is retrieved. The current output time parameter of the unified clock source is used as the matching benchmark to locate the node that completely corresponds to the time parameter on the timeline, thereby obtaining the 3D spatial data slice bound to that node. To ensure the reliability of subsequent data processing, the effective data unit ratio of the data slice needs to be checked simultaneously. This ratio is calculated by the ratio of the number of effective data units in the slice to the total number of data units. Only when this ratio is not less than 90% (which can be adjusted according to the safety level requirements of mountain operations, with a maximum of 95%) is the slice confirmed as a valid processing object. If the ratio is lower than the threshold, the data completion process is immediately initiated. Based on the same type of data from two adjacent effective sampling times before and after the slice, linear interpolation is used to supplement the missing data. The time interval is used as the weighting coefficient in the interpolation calculation to ensure that the deviation between the completed data and the actual working conditions is controlled within 5%. At the same time, the interpolation completion attribute is marked on the completed data to distinguish it from the original valid data.

[0032] Subsequently, according to the preset spatial grid indexing rules, all spatial grid cells within the valid data slice are traversed. During the traversal, differentiated coordinate extraction operations are performed for different types of sensor data: for tether connection point data, its mapped coordinates (Xt, Yt, Zt) in the global spatial reference system are directly extracted, and these coordinates have already undergone load attitude correlation correction in step 22; for load swing state data, since it is strongly correlated with the instantaneous center of gravity of the load, the origin coordinates (0, 0, 0) of the global spatial reference system are directly extracted as its spatial position identifier; for UAV-related data, the real-time position coordinates of each UAV in the global spatial reference system are extracted ( , , The coordinates have undergone conversion from the geodetic coordinate system to the global reference system and attitude correction. For environmental disturbance data, the corresponding spatial influence source coordinates are extracted based on the deployment location of the measuring equipment. If deployed on a navigation drone, the drone's coordinates are used; if deployed on a payload, the origin coordinates are used; and if deployed on a ground base station, the converted global reference system coordinates are used. All extracted coordinate values ​​must be calibrated against the 0.1m×0.1m×0.1m grid precision of the global spatial reference system. If the coordinates have insufficient decimal places, they are rounded to one decimal place to ensure consistent coordinate accuracy.

[0033] Next, a one-to-one correspondence between spatial coordinates and equilibrium state observations is established, quantifying the sensor data associated with each spatial coordinate point into equilibrium state observations. The specific association rules are as follows: the observation value corresponding to the tether connection point coordinates is the instantaneous tension of the tether, with an additional tension change rate parameter. This rate of change is calculated by dividing the tension difference between the current sampling time and the previous sampling time by the sampling period; the observation value corresponding to the load origin coordinates is the comprehensive quantized value of the load oscillation. The calculation uses the load's pitch angle, roll angle, and yaw angle as core parameters, combined with the angular velocity and angular acceleration corresponding to each angle, and sets weights according to the load mass distribution characteristics. The attitude parameters corresponding to the mass concentration area have their weight increased by 20%-30%, and the comprehensive quantized value is obtained through weighted summation. (UAV) The observed values ​​corresponding to the position coordinates are the attitude stability values ​​of the UAV, with the deviation of the roll angle, pitch angle and preset stable attitude as the core, combined with the fluctuation amplitude of the three-axis acceleration, and are converted into quantized values ​​of 0-10 through normalization. The larger the value, the more stable the attitude. The observed values ​​corresponding to the coordinates of the environmental disturbance source are the comprehensive disturbance intensity values, with wind speed as the basic parameter, gust intensity weighted at 1.5 times and airflow change rate weighted at 1.2 times. In high-altitude operation scenarios (altitude exceeding 1000m), the gust intensity weight is increased by an additional 0.3 times to ensure compliance with the airflow characteristics of high-altitude mountainous areas.

[0034] Finally, data cleaning is performed on all coordinate-observation pairs: data pairs whose observations exceed the physically reasonable range are removed, such as tether tension exceeding the sensor's maximum range (typically 1.5 times the drone's rated load) or load swing angle exceeding 60° (the physical limit angle); data pairs whose coordinates exceed the lifting operation space are also removed, with the load origin centered on the maximum lifting radius of the drone. After cleaning, the remaining data pairs are encapsulated according to the structure of spatial three-dimensional coordinates, equilibrium state observations, data type, timestamp, and validity identifier, forming a labeled spatial point set to provide standardized input for subsequent convex hull calculations.

[0035] Step 32: Using the coordinates of all spatial locations in the marked point set as input, perform 3D convex hull calculation to generate a minimum convex hull polyhedron that encloses all input points. The vertices of the minimum convex hull polyhedron are a subset of the point set, and its surface is composed of multiple triangular facets. Each triangular facet inherits the equilibrium state observations associated with its vertices, specifically including: The labeled spatial point set output in step 31 is preprocessed as follows: A coordinate deduplication algorithm is used to traverse all elements of the point set. If points with identical spatial coordinates exist, the point with higher observation accuracy is retained (judged by data validity indicators, with original valid data taking precedence over interpolated data). The remaining duplicate points are removed. Subsequently, a local outlier algorithm is used to detect outliers. Based on the spatial density of each point in the point set, an outlier threshold of 1.8 is set (which can be adjusted according to the point set density). Points with outliers exceeding the threshold are identified as potential anomalies and temporarily stored in an independent dataset. The remaining points are used as the core valid point set for convex hull calculation. After preprocessing, the total number of core point sets, the maximum and minimum spatial coordinate values ​​are recorded to clarify the spatial distribution range of the point set.

[0036] Next, the divide-and-conquer method is used to perform 3D convex hull calculation. Using all spatial coordinates of the core effective point set as input, the minimum convex hull polyhedron that encloses all input points is constructed. The specific calculation process is divided into three stages: The first stage is the splitting stage, which uses the X-axis component of the spatial coordinate as the reference to evenly divide the core point set into two subsets. During the splitting, it is ensured that the difference in the number of points between the two subsets does not exceed 1. The splitting operation is repeated until the number of points in each subset does not exceed 5. At this point, the size of the subsets meets the requirements for fast solution of the basic convex hull. The second stage is the solution stage, which calculates the local convex hull of each small subset by enumeration, that is, finds the smallest triangle that can enclose the subset and determines the vertices and surface patches of the local convex hull. The third stage is the merging stage, which takes the local convex hulls of two adjacent subsets and removes the enclosed inner surface patches by finding the common tangent of the two convex hulls. The common tangent must meet two conditions: first, the tangent is tangent to both local convex hulls at the same time; second, all vertices of the two convex hulls are located on the same side of the tangent. All local convex hulls are gradually merged in this way, and finally a global minimum convex hull polyhedron containing all points of the core effective point set is formed. The geometric features of this polyhedron are that it has a minimum volume, and all core points are located inside or on its surface. Its surface is composed of multiple non-overlapping triangular facets.

[0037] Subsequently, the inheritance and assignment of equilibrium state observations on the convex hull polyhedron are completed. Since the vertices of the polyhedron directly originate from the core valid point set, the equilibrium state observation, data type, validity identifier, and other attributes corresponding to each vertex are directly inherited into the vertex attributes. For each triangular facet on the polyhedron surface, its observations are calculated using a weighted average method: first, the spatial coordinates of the three vertices of the facet are determined, and the centroid coordinates of the facet are calculated (obtained from the average of the three vertex coordinates); the straight-line distances from each of the three vertices to the centroid are calculated, and the reciprocal of the distance is used as the weight of each vertex observation (the closer the distance, the greater the weight); the comprehensive equilibrium state observation of the triangular facet is calculated using the formula "(Vertex A observation × weight A + Vertex B observation × weight B + Vertex C observation × weight C) / (weight A + weight B + weight C)". Simultaneously, a traceability identifier is added to each triangular facet to clearly record which three vertices it comprises and the original data information of each vertex, ensuring the traceability of the observations.

[0038] Finally, the accuracy of the convex hull polyhedron is verified using a temporarily stored dataset of potential outliers: the shortest distance from each outlier to the polyhedron surface is calculated, determined by the minimum perpendicular distance between the outlier and each facet of the polyhedron. If the shortest distance is less than 0.2m (allowable error for operational accuracy), the outlier is considered to be within the reasonable coverage of the convex hull, and its observation is used as reference data and associated with the nearest facet. If the shortest distance exceeds 0.2m, the outlier is added to the core valid point set, and the above convex hull calculation process is repeated until all outliers meet the accuracy requirements, ultimately determining the shape of the minimum convex hull polyhedron and the observation attributes of each component.

[0039] Step 33: Based on the minimum convex hull polyhedron, perform topological subdivision within each triangular facet, insert new vertices, and connect them to form a dense triangular mesh; perform iterative smoothing filtering on the dense triangular mesh to eliminate the sharp edges formed by the original convex hull vertices, ultimately generating a continuous balanced surface, specifically including: Based on the minimum convex hull polyhedron determined in step 32, topological subdivision is performed on all triangular facets on its surface to increase the mesh density and meet the surface continuity requirements. Before subdivision, a precision standard is set: based on the spatial scale of the hoisting system (typically a drone hoisting radius of 5-10m), the maximum side length threshold for the subdivided triangular facets is set to 0.4m (adjustable within the range of 0.3m-0.5m). The subdivision operation rules are as follows: for each original triangular facet, connect the midpoints of its three sides to divide the original facet into four congruent sub-triangular facets. At this point, the side length of the sub-facet is half the side length of the original facet. For each sub-facet, check if the side length meets the threshold requirement. If it still exceeds the threshold, repeat the midpoint connection operation until the side length of all sub-facets does not exceed 0.4m. During the subdivision process, the newly inserted midpoint (i.e., the vertex of the sub-patch) needs to complete the assignment of coordinates and observation values: the midpoint coordinates are the average of the coordinates of the two endpoints of the corresponding edge; the equilibrium state observation value of the midpoint is calculated using linear interpolation, that is, the weight is allocated according to the endpoint observation value and the distance ratio between the midpoint and the endpoint. The closer to the endpoint, the closer the weight is to the endpoint observation value.

[0040] After subdivision, an initial triangular mesh is formed, consisting of numerous dense sub-triangular patches. This mesh fully retains the spatial enclosure characteristics of the original convex hull polyhedron, but the surface still contains sharp edges formed by the vertices of the original convex hull. To eliminate these sharp edges and make the mesh surface more continuous, a Laplacian smoothing filter is applied to the initial triangular mesh. The filtering operation is performed on each vertex in the mesh: first, all adjacent vertices of the vertex are identified (i.e., vertices directly connected by edges), and the number and coordinates of adjacent vertices are counted; the average spatial coordinates of adjacent vertices are calculated, and this average is used as the target smoothed coordinates of the vertex; the actual coordinates of the current vertex are updated to the original coordinates × (1-λ) + target smoothed coordinates × λ, where λ is the smoothing factor, with a value of 0.2 (ranging from 0.1 to 0.3; a larger value results in a more pronounced smoothing effect but can easily lead to shape distortion). It should be noted that the equilibrium state observations of the vertices do not participate in coordinate smoothing; they are only synchronously bound to the new position along with the coordinate update to ensure that the system equilibrium state information is not lost.

[0041] To avoid surface distortion caused by excessive smoothing (such as masking key imbalance features like slight load swaying), a dual iteration termination condition is set: first, a convergence condition, where the coordinate change of all vertices is calculated after each smoothing step, and the surface shape is considered stable when the maximum coordinate change is less than 0.01m; second, an iteration limit, where the maximum number of iterations is set to 6 based on the mesh density (the higher the mesh density, the more iterations can be), to avoid infinite iteration. During iteration, the topological integrity of the triangular mesh needs to be checked in real time: by determining whether the edges of adjacent faces correspond uniquely, ensuring no vertex overlaps, face intersections, or other anomalies; and by calculating the face normals, ensuring no flipped faces (normal vector direction abruptly exceeding 90°). If a topological anomaly occurs, the process immediately backtracks to the previous valid iteration result, reduces the smoothing factor by 0.05, and re-executes the filtering operation.

[0042] After smoothing, surface optimization is performed on the final triangular mesh: a mesh splicing check algorithm is used to traverse all sub-facets to ensure that adjacent facets completely overlap at the connection edges, without gaps or overlaps, forming a complete closed surface; local concavity and convexity defects on the surface are corrected by calculating the angle between the normal vectors of adjacent facets. If the angle exceeds 90°, it is judged as a concavity or convexity defect. The orientation of the defective facet is adjusted based on the average normal vector of the 10 adjacent facets around the defect area, so that the angle between the normal vectors is restored to a reasonable range (less than 30°); finally, the equilibrium state observations of each facet are mapped to the surface according to their spatial positions, forming a continuous equilibrium surface that combines spatial geometry and equilibrium state information. The core advantage of this surface is that the surface is continuous without obvious sharp edges, which can intuitively reflect the overall equilibrium status of the system, and the distribution of observations in each region of the surface accurately corresponds to the actual state of the UAV, the load, and the environment.

[0043] In a preferred embodiment of the present invention, step 4, determining the vertical projection points of the four UAV center-of-gravity positions on the continuous equilibrium surface, includes: Step 41: Extract the spatial position coordinates of each UAV from the three-dimensional spatial data array, as the actual position of the center of gravity of each UAV in three-dimensional space, specifically including: Using the time parameters used in step 31 for locating the 3D spatial data slice as a reference, the current sampling time is locked on the time axis of the 3D spatial data array, and the complete data slice corresponding to that time is retrieved. Since the identifiers of UAVs in the multi-UAV hoisting system are unique, based on a preset UAV number sequence (e.g., UAV-1 to UAV-4), the third data (UAV coordinate-attitude parameter data) within the data slice is classified and filtered to extract the preliminary spatial position coordinates corresponding to each UAV. , , The preliminary value is the coordinate result after the geodetic coordinate system to global spatial reference system transformation is completed in step 22.

[0044] Considering that changes in the drone's flight attitude will cause its center of gravity to shift relative to its geometric center, it is necessary to calibrate the initial coordinates using the drone's flight attitude data. Specifically, the drone's pitch angle α, roll angle β, and fuselage geometric parameters (including fuselage length, width, height, and a preset offset coefficient of the fuselage center of gravity relative to the geometric center) are extracted simultaneously from the third data. A local coordinate system is established with the drone's geometric center as the origin, with the X-axis along the direction of flight, the Y-axis along the lateral direction, and the Z-axis along the vertical direction. Based on the pitch angle α and roll angle β, the center of gravity offset (preset value) in the local coordinate system is transformed to the global spatial reference system to obtain the center of gravity offset coordinates (△X, △Y, △Z).

[0045] The initial coordinate values ​​are vector-superimposed with the centroid offset coordinates, i.e. , , The actual position coordinates of the center of gravity of each UAV in three-dimensional space were calculated. , , After calibration, the coordinates are validated: it is determined whether the coordinate values ​​are within a reasonable range of the global spatial reference system (a spherical region centered on the load origin with the maximum lifting radius of the UAV as its radius). If they exceed the range, the attitude data from the previous sampling time is extracted for secondary calibration until the coordinates meet the requirements. Finally, the number of each UAV is associated with and stored with the corresponding actual coordinates of its center of gravity, forming a mapping table of UAV number and center of gravity coordinates.

[0046] Step 42: Based on the attitude data and gravitational acceleration information corresponding to the spatial position coordinates of each UAV, calculate the vertical projection direction perpendicular to the local horizontal plane, specifically including: Acquire the real-time gravity acceleration vector data (gx, gy, gz) output by the inertial measurement unit (IMU) in the flight control system of each UAV. This vector represents the direction of the local gravity field measured by the UAV in its current flight attitude. It points perpendicular to the local horizontal plane and towards the Earth's center, serving as the core basis for determining the vertical projection direction. Since this vector is based on the UAV's local coordinate system, it needs to be combined with the UAV's pitch angle α, roll angle β, and yaw angle γ extracted in step 41 to transform it into a gravity acceleration vector (Gx, Gy, Gz) in the global space reference frame.

[0047] The coordinate transformation process must follow the rules of the three-dimensional space rotation matrix. First, the local coordinate system of the fuselage is rotated around the global Z-axis by the yaw angle γ to eliminate heading deviation. Then, the pitch angle is rotated around the rotated Y-axis to eliminate pitch deviation. Finally, the roll angle is rotated around the rotated X-axis to completely align the local coordinate system with the global space reference system, thus completing the coordinate transformation of the gravitational acceleration vector. The opposite direction of the transformed vector (Gx, Gy, Gz) is the upward direction perpendicular to the local horizontal plane, and the vertical projection direction required in this step is the positive direction of this vector, that is, the direction pointing to the local horizontal plane.

[0048] To improve the accuracy of the projection direction, the gravitational acceleration vector needs to be corrected based on the mountainous terrain. Using positioning devices deployed on ground base stations or payloads, the altitude and slope data of the current work area are obtained. If the slope exceeds 15°, the converted gravitational acceleration vector is fine-tuned according to the slope direction, compensating for the vector components along the perpendicular direction of the slope. The compensation coefficient is set according to the slope value (the larger the slope, the larger the compensation coefficient, ranging from 1.0 to 1.2). After fine-tuning, the vector is normalized to obtain a vertical projection direction vector (dx, dy, dz) per unit length. The direction of this vector is the reference direction for subsequent projection operations, ensuring that the projection direction is always perpendicular to the actual local horizontal plane of the work area, rather than the theoretical horizontal plane.

[0049] Step 43: Project the actual position of each UAV's center of gravity in three-dimensional space onto the continuous equilibrium surface along the vertical projection direction, and calculate the spatial intersection point of each projection line with the continuous equilibrium surface. Specifically, this includes: Based on the actual coordinates of the UAV's center of gravity (Xu', Yu', Zu') obtained in step 41 and the vertical projection direction vector (dx, dy, dz) obtained in step 42, the equation of the projection line for each UAV's center of gravity is constructed. The geometric definition of this projection line is: an infinitely long straight line extending along the vertical projection direction vector (dx, dy, dz) starting from the UAV's center of gravity, where all points in space satisfy the condition that the coordinates of any point = the coordinates of the starting point + the parameter t × the direction vector, where t is a real number greater than 0. The larger the value of t, the farther the point is from the UAV's center of gravity, and the longer it extends along the projection direction.

[0050] Since the continuous equilibrium surface is a closed surface composed of a large number of triangular facets (output of step 33), it is necessary to check the spatial relationship between the projection lines and each triangular facet to determine the intersection point. The checking process uses the ray-triangle intersection detection method. For each triangular facet, first obtain the spatial coordinates of its three vertices (P1, P2, P3) and the equation parameters of the plane on which the facet is located (calculated from the coordinates of the three points). Substitute the equation of the projection line into the equation of the plane on which the triangular facet is located to solve for the value of parameter t. If t is negative, it means that the intersection point is located in the opposite direction of the projection line (away from the side of the surface), and the facet is directly excluded; if t is positive, calculate the corresponding intersection point coordinates (X intersection, Y intersection, Z intersection).

[0051] After obtaining the intersection point coordinates, it is necessary to verify whether the intersection point is located within the interior region of the current triangular facet, rather than at any arbitrary position on the plane containing the facet. The verification method uses the core logic of the barycentric coordinate method: connect the three vertices of the triangular facet to the intersection point to form three sub-triangles, calculate the sum of the areas of these three sub-triangles, and if the difference between this sum and the area of ​​the original triangular facet is less than a preset precision threshold (usually 0.001m), then the verification is successful. 2 If the difference exceeds the threshold, the intersection point is determined to be inside the surface, which is a valid intersection point between the projection line and the continuous equilibrium surface; if the difference exceeds the threshold, the surface is excluded and the next surface is detected.

[0052] If the projection line intersects with multiple triangular patches (due to local overlap of the curved surface in special cases), the straight-line distance between each intersection point and the starting point of the UAV's center of gravity is calculated. The intersection point with the smallest distance is selected as the final spatial intersection point. This intersection point is the closest point of the UAV's center of gravity projected onto the curved surface in the vertical direction, which conforms to the definition of geometric projection. If no valid intersection point is detected, it indicates that there is a deviation in the projection direction. Step 42 of the projection direction calculation needs to be re-executed, the terrain compensation coefficient is adjusted, and the detection is repeated until a valid intersection point is obtained.

[0053] Step 44: Record the spatial intersection points of each projection line and the continuous equilibrium surface as the vertical projection points of the corresponding UAV's center of gravity on the continuous equilibrium surface. Specifically, this includes: The final spatial intersection coordinates (X intersection, Y intersection, Z intersection) calculated in step 43 are defined as the vertical projection points of the corresponding UAV's center of gravity on the continuous equilibrium surface. An association mapping relationship is established between the UAV number, the actual position of the center of gravity, and the coordinates of the projection point to ensure that each projection point can be accurately traced back to its corresponding UAV and original position data.

[0054] To ensure the integrity and traceability of the projection point data, it is necessary to supplement the associated attribute information for each projection point, including: the precise timestamp of data acquisition (consistent with the sampling time of the 3D spatial data array), the direction vector parameters of the projection line (dx, dy, dz), the identification information of the triangular facet where the intersection point is located (the unique number assigned to the facet in step 33), and the validity identifier of the intersection point detection process (such as single intersection point, selecting the nearest intersection point among multiple intersection points, etc.). This attribute information will provide basic data support for the construction of the four-boundary feature configuration region in subsequent steps, and at the same time facilitate subsequent fault diagnosis and data traceability.

[0055] The projection point data of all four UAVs are aggregated to form a standardized projection point dataset containing UAV number, centroid coordinates, projection point coordinates, direction vector, patch identifier, timestamp, and validity identifier. After aggregation, a data consistency check is performed: it is checked whether all four projection points are located on the outer surface of the continuous equilibrium surface (determined by the direction of the patch's normal vector; if the normal vector points outward, it is on the outer surface). If projection points exist on the inner surface, the intersection detection in step 43 is re-executed to eliminate interference from patches inside the surface. After successful verification, the dataset is stored in a temporary cache, providing direct input for the construction of the four-boundary feature configuration region in step 5.

[0056] This step provides a high-quality spatial benchmark for imbalance early warning through precise coordinate extraction, terrain-adaptive projection, and rigorous data processing, yielding significant benefits: First, accurate center of gravity positioning: Step 41 combines UAV attitude data to calibrate coordinates, eliminating interference from center of gravity shifts caused by changes in aircraft attitude. Validation ensures the coordinates are within a reasonable operating range, providing a reliable starting point for projection. Second, strong adaptability to projection direction: Step 42 converts gravity vectors based on IMU data and combines it with fine-tuning compensation for mountainous terrain slopes, ensuring the projection direction is always perpendicular to the actual horizontal plane, overcoming the limitations of the theoretical horizontal plane. Third, reliable projection point positioning: Step 43 accurately locks surface intersections through ray-triangle detection and intersection verification, avoiding positioning errors caused by surface overlap or directional deviations. Fourth, strong data support: Step 44 constructs a standardized dataset with complete associated attributes, and invalid data is eliminated through consistency checks, providing traceable and unbiased core input for subsequent four-boundary configuration region construction, enhancing the accuracy and reliability of imbalance analysis.

[0057] In a preferred embodiment of the present invention, step 5 involves constructing a four-boundary feature configuration region based on the vertical projection points; and solving for the geometric imbalance correction coefficient based on the intersection of the four-boundary feature configuration region and the continuous equilibrium surface feature boundary. Step 51: Based on the spatial position of the vertical projection point on the continuous equilibrium surface, connect adjacent projection points sequentially according to the preset UAV spatial orientation order to construct a spatial quadrilateral on the continuous equilibrium surface. Define this spatial quadrilateral as a four-boundary feature configuration region. Specifically, this includes: retrieving the standardized projection point dataset output in step 44, extracting the coordinates of the vertical projection points corresponding to the four UAVs, and denoting them as P1(X1, Y1, Z1), P2(X2, Y2, Z2), P3(X3, Y3, Z3), and P4(X4, Y4, Z4), respectively. At the same time, obtain the triangular facet identification information of the intersection points associated with each projection point to clarify the specific position of each projection point on the continuous equilibrium surface.

[0058] Based on the preset operation layout of the multi-machine hoisting system, the spatial orientation sequence of the UAVs is determined. A unified rule of clockwise circling the load is adopted. With the instantaneous center of gravity of the load (the origin of the global spatial reference system) as the center, the azimuth angles of each projection point relative to the origin are calculated (with the positive X-axis direction as 0° and increasing counterclockwise). P1-P4 are sorted to obtain a fixed connection order. This order ensures that adjacent projection points are distributed in a circular pattern in space, avoiding cross connections that could lead to distortion of the configuration area.

[0059] Along the topological structure of the continuous equilibrium surface, adjacent projection points are connected sequentially to construct a spatial quadrilateral: For two adjacent points (e.g., P1 and P2), based on the identifier of their respective triangular facets, the shortest path connecting the two points is retrieved in the triangular mesh of the surface. The path consists of continuous triangular facet edges, and all points on the path lie on the surface. A mesh traversal algorithm ensures that the path conforms to the surface shape, rather than being a straight line connection that traverses the interior of the surface. After connecting all adjacent points in this way, a closed spatial polygon is formed, which is completely attached to the continuous equilibrium surface and is defined as a quad-boundary feature configuration region.

[0060] The validity of the constructed spatial quadrilateral is verified by calculating the sum of its interior angles and the distribution of its side lengths. If the deviation between the sum of its interior angles and the sum of the interior angles of the planar quadrilateral (360°) is within 5°, and the ratio of any two side lengths does not exceed 2 (meeting the balanced layout requirements for multi-machine hoisting), the configuration region is deemed valid. If the deviation exceeds the standard, the order of projection points and connection paths are rechecked to eliminate connection errors caused by local concavities and convexities of the curved surface until a valid region is formed. Finally, the coordinates of the boundary vertices of this region, the key nodes on the connection paths, and the information of the triangular facets they belong to are recorded to provide basic data for subsequent steps.

[0061] Step 52: Based on the spatial distribution of all data points on the continuous equilibrium surface, extract the outer contour boundary line of the continuous equilibrium surface as its feature boundary. Specifically, this includes: obtaining the complete triangular mesh data of the continuous equilibrium surface generated in Step 33, which contains the vertex coordinates, facet normals, and relationships between adjacent faces of all triangular faces. Traverse all vertices in the mesh, calculate the edge attributes of each vertex, and count the number of triangular faces sharing that vertex. If the count is 1, the vertex is an edge vertex of the surface (belonging to only one facet and located at the surface boundary); if the count is ≥2, it is an internal vertex. Based on the set of edge vertices, perform edge connection to extract the outer contour boundary line: randomly select an edge vertex as the starting point, find the edge edge contained only by one facet according to the edge information of its triangular facet, and locate the next edge vertex along that edge edge; starting from the new vertex, repeat the edge edge retrieval and vertex location operations to form a continuous vertex sequence. When the sequence returns to the starting vertex, the extraction of a closed contour line is complete.

[0062] Since the continuous equilibrium surface is a closed convex surface that encloses all sensor data points, there is only one external contour boundary line. The extracted contour line is optimized by: removing redundant vertices (repeated vertices with a spacing of less than 0.05m between adjacent vertices) caused by mesh subdivision; eliminating local jaggedness of the contour line through smoothing; calculating the spatial extent of the contour line to ensure it completely encloses the four-boundary feature configuration region constructed in step 51; if there are local incomplete enclosement cases, edge vertex retrieval is performed again to supplement missing boundary segments. Finally, the optimized contour line is defined as the feature boundary of the continuous equilibrium surface, and its vertex coordinate sequence and the connection relationship between adjacent vertices are recorded.

[0063] Step 53: Calculate the spatial intersection points of the edges of the four-boundary feature configuration region with the feature boundary lines of the continuous equilibrium surface to obtain the feature boundary intersection points. Specifically, this includes defining the four edges of the four-boundary feature configuration region as line segments L1(P1-P2), L2(P2-P3), L3(P3-P4), and L4(P4-P1), respectively. Each line segment is a spatial curve formed by connecting along the surface topology path in Step 51. Digital representation is achieved through the coordinate sequence of key nodes on the path (e.g., the node sequence of L1 is P1, A1, A2, ..., P2). Simultaneously, the feature boundary lines extracted in Step 52 are represented in the form of a vertex sequence to clarify the direction of their spatial curves.

[0064] For each region boundary line (e.g., L1) and feature boundary line, spatial curve intersection detection is performed: using a piecewise approximation method, both curves are divided into several short straight line segments at 0.1m intervals (e.g., L1 is divided into P1-A1, A1-A2, etc., and the feature boundary line is divided into Q1-Q2, Q2-Q3, etc.); for each pair of short straight line segments, the existence of an intersection point is determined by judging whether they are coplanar and whether their endpoints cross regions. If an intersection point exists, the precise coordinates of the intersection point are calculated using linear interpolation to ensure that the coordinate accuracy is consistent with the surface mesh accuracy (0.1m).

[0065] The detected intersections are filtered for validity: only intersections that simultaneously satisfy the condition of being located within a valid segment of the region boundary line and a valid segment of the feature boundary line are retained, i.e., the intersection is between the two endpoints of line segment L and between adjacent vertices of the feature boundary line. If a region boundary line (e.g., L2) has no intersection with a feature boundary line, it means that the edge is completely inside the surface and is recorded as having no intersection. If multiple intersections exist, the two intersections closest to the region vertices are selected as feature boundary intersections (to avoid interference from multiple intersections caused by local surface wrinkles). All valid intersections are summarized to form a set of feature boundary intersections. Each intersection is associated with its region boundary line identifier (e.g., L1-L4) and feature boundary line segment identifier, and the spatial coordinates and triangular facet information of the intersection are recorded to provide a basis for subsequent closed region division.

[0066] Step 54: Using the continuous equilibrium surface as the top surface and the plane passing through the four vertices of the four-boundary feature configuration region and parallel to the horizontal plane as the bottom surface, calculate the spatial volume between the continuous equilibrium surface and the bottom surface, which serves as the first reference volume. Specifically, this includes: Determine the spatial location of the horizontal base, extract the Z coordinates (Z1-Z4) of the four vertices P1-P4 of the four boundary feature configuration region, and calculate their average value Z. 平 , with Z=Z 平 For the height, construct a plane parallel to the local horizontal plane (determined by the vertical projection direction in step 42). This plane is the horizontal base of the calculated volume, and its equation is determined based on the global spatial reference frame, ensuring that the plane is perpendicular to the vertical projection direction.

[0067] Define the spatial scope for volume calculation: with the continuous equilibrium surface as the top surface and the horizontal bottom surface as the bottom surface, the volume calculation region is all surfaces between the two surfaces that satisfy Z≥Z 平 (If the curved surface is above the bottom surface) or Z≤Z 平 (If the curved surface is below the bottom surface) spatial region. Since the continuous equilibrium curved surface is a convex surface that encloses the load and the UAV, the top and bottom surfaces form a closed spatial geometry.

[0068] The first reference volume is calculated using a mesh partitioning integration method: the triangular mesh of the continuous equilibrium surface is layered according to the height of the horizontal base. For each triangular facet, its positional relationship with the horizontal base is determined. If the facet is completely above the base, the volume of the frustum formed by the facet and the base in the vertical direction is directly calculated. If the facet partially spans the base, it is divided into two parts, one above and one below the base, and only the volume corresponding to the upper part is calculated. If the facet is completely below the base, the volume is not included. The volume components corresponding to all triangular faces are summed to obtain the total spatial volume between the continuous equilibrium surface and the horizontal base, which is defined as the first reference volume V1. During the calculation, the orientation of the facets is determined by the normal vector direction of the triangular faces to ensure that the volume calculation direction is consistent with the convexity of the surface and to avoid negative volume values. At the same time, the volume contribution value of each facet is recorded to provide support for subsequent data traceability.

[0069] Step 55: Based on the intersection points of the feature boundaries, a closed surface region is defined on the continuous equilibrium surface, jointly bounded by the four-boundary feature configuration region and the surface feature boundaries. The volume of space between the closed surface region and the horizontal bottom surface is calculated as the second actual volume. Specifically, based on the set of feature boundary intersection points from Step 53, the closed surface region is defined on the continuous equilibrium surface: For each edge of the four-boundary feature configuration region, if there is a feature boundary intersection point (such as intersection points C1 and C2 on L1), the intersection point replaces the original region vertex to form a new boundary node; if an edge has no intersection point (such as L2), the original vertex P2 is retained. Following the original region's encirclement order, the new boundary nodes are connected to the corresponding segments on the feature boundaries to form a closed surface region enclosed by the region boundary segments and the feature boundary segments. This region is the actual effective coverage area of ​​the four-boundary feature configuration region on the continuous equilibrium surface, excluding invalid space outside the surface.

[0070] Determine the set of triangular faces corresponding to the closed region: By retrieving the triangular facet identifiers contained within the boundary of the closed region and combining the topological relationships of the surface mesh, filter out all triangular faces completely located inside the closed region, as well as triangular faces partially located within the region (which need to be segmented). For faces partially located within the region, segment them into internal sub-facets and external sub-facets based on the position of the region boundary line, retaining only the internal sub-facets for volume calculation.

[0071] Using the horizontal bottom surface from step 54 (Z=Z) 平 As the base for volume calculation, the same mesh partitioning integration method as the first reference volume is used to calculate the spatial volume between the closed curved surface region and the horizontal base: for the selected internal triangular facets, the volume of the frustum formed by them and the base or the volume components after division are calculated one by one to ensure the consistency of the calculation method with V1 and eliminate calculation errors.

[0072] Summing the volume components of all internal facets yields the spatial volume between the closed curved surface region and the horizontal base, defined as the second actual volume V2. After calculation, the numerical ranges of V2 and V1 are compared. If V2 > V1, it indicates an error in the region division, requiring a re-examination of the validity of the feature boundary intersections and the region segmentation logic until V2 ≤ V1 and conforms to the actual spatial distribution.

[0073] Step 56: By comparing and analyzing the degree of difference between the first reference volume and the second actual volume, calculate the geometric imbalance correction coefficient. Specifically, this includes: clarifying the physical meaning of the first reference volume V1 and the second actual volume V2: V1 represents the ideal spatial volume benchmark corresponding to the system equilibrium state when the four UAVs are subjected to equal forces; V2 represents the actual spatial volume occupied by the four boundary feature regions under the current working conditions. The difference between the two directly reflects the degree of deviation of the system's spatial equilibrium state. The greater the difference, the more unbalanced the forces on the UAVs are, and the higher the risk of system imbalance.

[0074] Calculate the basic imbalance difference value and difference rate: The basic difference value ΔV = |V1 - V2|, and the difference rate η = ΔV / V1 × 100%. This difference rate quantifies the deviation ratio of the actual volume from the ideal volume, providing a basic indicator for coefficient calculation. To avoid distortion of the difference rate due to an excessively small V1, when V1 < a preset minimum volume threshold (set according to the load volume, usually 1.5 times the load volume), the load volume is used instead of V1 for difference rate calculation.

[0075] The weighted adjustment of the difference rate is based on the characteristics of the load type: Different power emergency repair loads (such as transformers, cable reels, and insulators) have different mass distributions and aerodynamic characteristics, resulting in varying sensitivities to imbalance. The type characteristic parameter table of the current hoisting load is retrieved to obtain the imbalance sensitivity coefficient k corresponding to that load (k≥1, the higher the sensitivity, the larger the k value, e.g., k=1.8 for transformers, k=1.2 for insulators); the difference rate η is multiplied by the sensitivity coefficient k to obtain the corrected difference rate η' = η×k.

[0076] Define a geometric imbalance correction coefficient K, with a value range of 0-10. The calculation logic is as follows: when η' < 5% (the system is basically in balance), K = η' × 2 (the coefficient is linearly related to the difference rate and the value is small); when 5% ≤ η' ≤ 30% (the system has a slight to moderate imbalance), K = 1 + (η' - 5%) × 0.4 (the coefficient increases faster, strengthening the sensitive response to imbalance); when η' > 30% (the system is severely imbalanced), K = 10 (the coefficient is capped, triggering a high-level warning).

[0077] The calculated K value is calibrated: combining the correspondence between the correction coefficient and the actual imbalance state in the historical hoisting data, if the current K value deviates from the historical coefficient under similar working conditions by more than 20%, the K value is finely adjusted according to the historical data to ensure that the coefficient can accurately reflect the actual imbalance state of the system; finally, the calibrated geometric imbalance correction coefficient K is output, and the parameters such as V1, V2, η, and k in its calculation process are associated to form a complete coefficient calculation file.

[0078] This step provides precise support for load imbalance assessment through multi-dimensional spatial analysis and quantitative calculation. The core benefits are as follows: First, the configuration region is constructed accurately and reliably. Based on the surrounding sorting of UAV projection points and the connection of curved surfaces, cross-distortion is avoided. Combined with interior angle sum and side length verification, the validity of the region is ensured, locking in the core spatial range for imbalance analysis. Second, feature boundary and intersection point extraction eliminates interference. The optimized curved surface contour completely surrounds the configuration region, and the intersection points detected by the piecewise approximation method are effectively screened, ensuring the accuracy of subsequent region division. Third, the volume difference quantification of imbalance is scientific and intuitive. The comparison between reference volume and actual volume, combined with load sensitivity coefficient correction, eliminates the influence of load characteristic differences. Historical data calibration further improves the accuracy of the coefficients. Fourth, the output geometric imbalance correction coefficient is both quantitative and traceable, providing core indicators for the comprehensive feature vector and clear data for subsequent early warning and fault diagnosis, significantly enhancing the scientific rigor and reliability of the imbalance assessment.

[0079] In a preferred embodiment of the present invention, step 6, generating a comprehensive load imbalance feature vector based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the currently hoisted power repair load type, includes: Step 61: Based on the geometric imbalance correction coefficient, use it as a scalar feature to characterize the imbalance state of the four-boundary feature configuration region. Specifically, this includes: retrieving the geometric imbalance correction coefficient output in step 56, and extracting the calibrated final geometric imbalance correction coefficient K value. This value is the core quantitative indicator of the spatial imbalance state of the four-boundary feature configuration region. The value range is stable between 0 and 10. The larger the value, the higher the risk of imbalance in the corresponding region.

[0080] To enhance the traceability and relevance of scalar features, the K value is bound to key parameters in the calculation process, forming a scalar feature data package. The bound parameters include: the first reference volume V1, the second actual volume V2, the basic difference value ΔV, the corrected difference rate η', and the load sensitivity coefficient k. These parameters explain the generation logic of the K value from different dimensions, providing background support for subsequent abnormal pattern recognition.

[0081] The scalar feature is time-aligned: the sampling time timestamp associated with the K value calculation is extracted and matched and calibrated with the unified time reference of the multi-source sensing data to ensure that the time attribute of the scalar feature is completely synchronized with the subsequently extracted time-frequency domain features, and the time deviation is controlled within 1 / 10 of the sampling period (i.e., 5ms-20ms) to avoid feature fusion distortion due to time misalignment.

[0082] Finally, the K value after binding parameters is defined as a scalar feature characterizing the imbalance state of the four-boundary feature configuration region. Its physical meaning is clarified as a macroscopic quantitative indicator of the spatial balance state of the multi-machine hoisting system. It is stored in the structure of scalar feature value-associative parameter set-time stamp to prepare for feature splicing in step 64.

[0083] Step 62: Extract key time-domain and frequency-domain features from multi-source sensor data. Specifically, this includes: using the timestamp of the scalar features in Step 61 as a reference, locking sensor data at the same sampling time and two sampling periods before and after (a total of 5 consecutive time points) in the multi-source sensor data packet to form a continuous data stream within a time window. This time window can reflect the real-time status of the data and capture short-term dynamic trends. The window length can be adjusted according to the system response speed (usually 250ms-1000ms).

[0084] Temporal feature extraction is performed as follows: For the tether tension data of each UAV, the real-time tension value at each moment is extracted, and the mean (reflecting the stability level of tension), fluctuation variance (reflecting the intensity of tension fluctuation), and the difference between the maximum and minimum values ​​(reflecting the extreme changes in tension) within the time window are calculated; For the load swing state data, the instantaneous values ​​of the three-dimensional angular velocity and angular acceleration at each moment are directly extracted, and the rate of change of each parameter within the time window (i.e., the difference between adjacent moments) is calculated; For the flight attitude data of each UAV, the instantaneous values ​​of the pitch angle and roll angle at each moment are extracted, and the deviation value and deviation variance from the preset stable attitude angle are calculated; For the mountainous environmental disturbance data, the instantaneous wind speed value and gust intensity at each moment are extracted, and the peak value and occurrence time of the gust intensity are recorded.

[0085] Frequency domain feature extraction: The continuous data stream within the time window is transformed into the frequency domain using the Fast Fourier Transform (FFT) method. The following frequency domain features are extracted: the dominant frequency of the tether tension signal of each UAV (reflecting the dominant frequency of tension fluctuation; if the dominant frequency is close to the load resonance frequency, it needs to be focused on) and the amplitude corresponding to the dominant frequency (reflecting the intensity of the dominant fluctuation); the frequency band energy distribution of the load swing angular velocity signal, dividing the frequency domain into three bands: low (0-1Hz), medium (1-5Hz), and high (5-10Hz), and calculating the proportion of energy in each band to the total energy (a high proportion of energy in the high-frequency band indicates severe load swing); and the spectral entropy of the environmental wind speed signal (reflecting the complexity of wind speed fluctuation; the larger the spectral entropy, the more unstable the airflow).

[0086] The extracted time-frequency domain features are cleaned and filtered: outliers caused by sensor noise are removed (using the 3σ criterion, i.e., values ​​exceeding the mean ± 3 standard deviations), and linear interpolation is used to supplement them with feature values ​​from adjacent time points; redundant features are removed through feature correlation analysis, such as retaining the feature set with the larger variance when the correlation coefficient of the pull variance of two drones exceeds 0.9, thus reducing feature dimensional redundancy. The final result is a key time-frequency domain feature set containing real-time values, statistical values, and frequency domain parameters, with each feature labeled with the corresponding sensor data type, drone number (if applicable), and timestamp.

[0087] Step 63: Based on the power repair load type specified in the current hoisting task, retrieve the static and dynamic characteristic parameters corresponding to the key time-domain and frequency-domain features from the preset load characteristic database. Specifically, this includes: constructing a preset load characteristic database. This database adopts a categorized index structure, with the power repair load type as the primary index (e.g., transformer, cable reel, insulator string, disconnector, etc.) and the load model as the secondary index. Each index entry stores complete static and dynamic characteristic parameters. The database supports real-time updates and can import characteristic parameters of new loads via ground base stations, ensuring that the parameters cover various common emergency repair loads.

[0088] Based on the current hoisting task's dispatch instructions, the specified power repair load type and model (e.g., "10kV oil-immersed transformer-S11-200kVA") are extracted from the task. This is then used as the search keyword to perform a precise matching query in the load characteristic database. If multiple similar entries exist in the search results (e.g., the same model load from different manufacturers), further filtering is performed based on the parameter information in the load's factory certificate of conformity to ensure unique matching.

[0089] The corresponding characteristic parameters are retrieved from the matched database entries. The static characteristic parameters include: the rated mass of the load (accurate to 0.1 kg, providing a basis for imbalance sensitivity analysis), the center of mass position coordinates (the offset relative to the geometric center of the load, in three-dimensional coordinate form, used to correlate load sway characteristics), and aerodynamic shape coefficients (including frontal area, drag coefficient, and lift coefficient, which are directly related to the degree of influence of environmental disturbances). The dynamic characteristic parameters include: the maximum allowable sway angle threshold of the load (set separately in the pitch, roll, and yaw directions; exceeding the threshold poses a safety risk), the natural vibration frequency of the load (used to compare with the dominant frequency in the frequency domain characteristics to determine whether there is a risk of resonance), and the load's inertial parameters (moment of inertia, which affects the response speed of load sway).

[0090] The validity of the invoked characteristic parameters is verified: the rated mass is compared with the total rated load of the UAV lifting system to ensure that the load mass does not exceed 80% of the system's rated load (ensuring system redundancy); the coordinates of the center of gravity are checked to ensure they are within the load geometry range, and the aerodynamic coefficients are within the typical range for similar loads (e.g., drag coefficients are typically between 0.8 and 1.5). If any parameters are abnormal, the database backup parameter retrieval mechanism is immediately triggered, retrieving the average characteristic parameters for this load type as temporary substitutes, marking the parameters as needing calibration, and simultaneously sending a parameter abnormality alert to the ground control terminal. After successful verification, the parameters are organized according to a static-dynamic classification structure to form a load characteristic parameter set.

[0091] Step 64: Normalize and splice the geometric imbalance correction coefficient, key time-domain and frequency-domain features, and static and dynamic characteristic parameters according to a preset feature order to form a comprehensive load imbalance feature vector. Key time-domain and frequency-domain features include: real-time values, mean values, and variances of tether tension for each UAV; angular velocity and angular acceleration of load swing; pitch and roll angles for each UAV; and instantaneous wind speed and gust intensity in mountainous environmental disturbances. Static and dynamic characteristic parameters include the load's rated mass, center of mass position, aerodynamic shape coefficient, and maximum allowable swing angle threshold. Specifically, this includes: determining the preset splicing order of the feature vector, which is set according to the logic of macro-micro-basic constraints, specifically: geometric imbalance correction coefficient (scalar feature), tether tension time-frequency domain features, load swing time-frequency domain features, UAV attitude time-frequency domain features, environmental disturbance time-frequency domain features, load static characteristic parameters, and load dynamic characteristic parameters. This order ensures the logical correlation within the feature vector, facilitating the feature weight allocation of the subsequent pattern recognition algorithm.

[0092] Standardization and normalization are performed on all features to uniformly map all feature values ​​to the [0,1] interval, eliminating the weight imbalance caused by differences in the units and numerical ranges of different features. Different normalization methods are used for different types of features: For features with a clear range of values ​​(such as geometric imbalance correction coefficient 0-10, pitch angle -30° to 30°), linear normalization is used, i.e., feature normalized value = (original feature value - feature minimum value) / (feature maximum value - feature minimum value); For features without a clear range but with a statistical benchmark (such as average tension value, instantaneous wind speed value), mean-standard deviation normalization is used. The mean and standard deviation are calculated based on the feature statistical data of similar historical hoisting tasks, and the feature normalized value = (original feature value - mean) / standard deviation. If historical data is insufficient, the statistical values ​​of the previous 100 sampling periods of the current task are used as the benchmark; For static parameters (such as rated mass, aerodynamic coefficient), relative normalization is used, with the upper limit of system design as the benchmark (such as rated mass with the upper limit of 1000 kg as the benchmark), and the relative proportion is calculated as the normalized value.

[0093] Normalized feature stitching is performed: Following a preset order, each normalized feature value from the scalar features, time-frequency domain feature set, and load characteristic parameter set is sequentially arranged to form a one-dimensional feature sequence. For example, the geometric imbalance correction coefficient K=5 (normalized to 0.5) is used as the first element of the sequence, followed by the normalized features of the four UAVs, such as real-time thrust, mean, variance, dominant frequency, and dominant frequency amplitude, until all features are arranged. During the stitching process, a unique feature identifier code is added to each feature value (e.g., K-001, thrust mean-UAV1-002), facilitating subsequent feature importance analysis and fault tracing.

[0094] Perform integrity and consistency checks on the spliced ​​feature sequence: count the total length of the feature sequence and compare it with the preset number of features (the preset number is determined according to the feature type, usually 50-80 dimensions) to ensure that no features are missing; check whether all feature values ​​are within the range of [0,1], and if there are values ​​outside the range, re-perform the normalization process of the feature to check for errors in the selection of the benchmark value; verify whether the timestamps of all features are completely consistent to ensure that the feature sequence reflects the system state at the same sampling time.

[0095] After successful verification, the one-dimensional feature sequence is formally defined as a comprehensive load imbalance feature vector. Each dimension of the vector corresponds to a specific feature value and an associated feature identifier. Simultaneously, metadata information for the feature vector is generated, including: the sampling timestamp of vector generation, the number of feature types involved in the concatenation, the baseline parameters used for normalization, and feature entries with anomaly annotations (such as temporarily substituted load parameters). Finally, the comprehensive load imbalance feature vector and metadata information are transmitted together to the collaborative anomaly pattern recognition module as core input data for load imbalance trend judgment, and simultaneously stored in the system database for at least 72 hours for subsequent analysis.

[0096] This step generates high-quality imbalance feature vectors through multi-dimensional feature fusion and standardization, providing core support for early warning with significant benefits: First, scalar features are reliably correlated, binding the geometric imbalance correction coefficient K with calculation parameters and aligning them temporally to ensure that macroscopic imbalance indicators are traceable and synchronized with sensor data, avoiding fusion distortion; Second, time-frequency domain features are accurate and effective, focusing on extracting real-time values, statistical values, and frequency domain parameters from core data, and using the 3σ criterion for noise reduction and correlation analysis for dimensionality reduction to retain key information on load and UAV dynamics; Third, load parameters have strong adaptability, accurately retrieving load characteristic parameters from the database, and verifying their reliability to adapt to the imbalance sensitivity characteristics of different emergency repair loads; Fourth, the feature vectors are of excellent quality, with normalization eliminating dimensional interference, and being assembled in logical order and rigorously verified to form vectors with reasonable dimensions and consistent data, providing comprehensive and accurate input for subsequent collaborative anomaly identification, and significantly improving the scientificity and reliability of imbalance trend judgment.

[0097] In a preferred embodiment of the present invention, step 7 involves performing collaborative abnormal pattern recognition on the comprehensive load imbalance feature vector based on the comprehensive load imbalance feature vector, identifying early trend characteristics of load imbalance, and generating a load imbalance fault early warning signal in real time when the load imbalance risk is detected to reach a preset warning level. This includes: Step 71: Based on the comprehensive load imbalance feature vector, identify isolated anomalies and combinations of abnormal features that deviate from the normal collaborative mode. Specifically, this includes constructing a normal collaborative mode library for multi-machine hoisting systems as a benchmark for anomaly identification. The data source for this mode library includes two parts: first, a set of comprehensive load imbalance feature vectors from historical similar hoisting tasks under unbalanced fault conditions, covering typical scenarios with different load types, environmental disturbance intensities, and UAV layouts; second, ideal balance condition feature vectors generated through simulation, simulating the system state when the load mass is uniform, the environment is windless, and the UAV attitude is stable. Cluster analysis is performed on the two types of data to form multiple normal collaborative mode cluster centers. Each cluster center corresponds to a typical normal operating condition. At the same time, the normal value range (mean ± 2 standard deviation) of each dimension of the feature vector under each type of mode is calculated.

[0098] Perform isolated outlier identification: Retrieve the current comprehensive load imbalance feature vector generated in step 64, and compare each dimension's feature value with the value range of the corresponding cluster centers in the normal pattern library. If a feature value in a certain dimension exceeds the value range of all normal cluster centers, and the importance weight of that feature (preset weights, with the geometric imbalance correction coefficient having the highest weight of 0.3, the tether tension fluctuation variance weight being the next highest at 0.2, and the weights of other features set according to correlation) is greater than 0.1, then that feature value is marked as an isolated outlier. For example, a geometric imbalance correction coefficient K=8 (normal range 0-4) and a tether tension fluctuation variance of a certain UAV exceeding the normal range by 3 times will both be identified as isolated outliers. At the same time, the feature identifier, original value, and deviation from the normal range corresponding to the outlier are recorded.

[0099] Anomaly feature combination identification is performed: Based on the collaborative association rules between features in the normal pattern library (extracted from historical data through association mining algorithms, such as a high average tether tension and a slightly larger corresponding drone pitch angle; increased wind speed and a simultaneous slight increase in load swing angular velocity), a feature collaborative association matrix is ​​constructed, where matrix elements represent the normal association strength between two features. All features in the current feature vector are subjected to pairwise association verification. If the association strength of a certain feature combination (such as a sudden increase in tether tension and no change in the corresponding drone pitch angle; an increase in load swing angular velocity and no significant change in ambient wind speed) is lower than the threshold in the normal association matrix (usually 50% of the normal strength), then this combination is determined to be an anomaly feature combination.

[0100] The identification results are deduplicated and validated: If multiple isolated outliers originate from the same physical cause (e.g., gusts causing abnormal wind speeds, which in turn lead to load swings and tension fluctuations), they are merged into a set of associated outliers. For combinations of anomalous features, the current environmental data (e.g., whether there are sudden gusts) is used to determine whether they are caused by environmental disturbances. If they are caused by environmental disturbances and the features can be recovered after the disturbances disappear, they are marked as temporary anomalous combinations; otherwise, they are marked as persistent anomalous combinations. The final output includes a set of isolated outliers (including association identifiers), a list of anomalous feature combinations (including temporary / persistent markers), and the corresponding original feature data.

[0101] Step 72: Based on the current time and the isolated anomalies and anomaly feature combinations from multiple consecutive previous times, determine the early trend characteristics of load imbalance. Specifically, this includes: constructing a time-series feature sequence: taking the current time as the termination node, extract the comprehensive load imbalance feature vectors and anomaly identification results from the previous N consecutive sampling times (N is set according to the system response speed, usually 10-20, corresponding to a time span of 500ms-4s), and form a time-series feature matrix in chronological order. Each row of the matrix corresponds to the feature vector of a time, and each column corresponds to a feature dimension. At the same time, an anomaly marker column for each time is added (recording isolated anomaly points and anomaly combination information).

[0102] Analyzing the continuity and evolution of anomalies: For isolated anomalies in the time-series feature matrix, the frequency of occurrence of each anomaly at consecutive time points is statistically analyzed. If an isolated anomaly of a certain feature appears at three or more consecutive time points, and the deviation shows an increasing trend (e.g., K value from 5, 6 to 7), then the anomaly is determined to be a trending anomaly. If it appears only at a single or discontinuous time point, and the deviation does not increase, then it is determined to be an occasional anomaly. For combinations of anomalous features, their transmissibility in time series is observed. If a tensile anomaly combination appears first, followed by a load sway anomaly combination and a UAV attitude anomaly combination appearing sequentially at consecutive time points, then an imbalance transmission chain is formed, which is determined to be a trending anomaly combination.

[0103] For trend anomalies, calculate the number of consecutive occurrences (number of times they occur), the maximum deviation (maximum difference from the normal range), and the average growth rate (ratio of the deviation increment to time). For trend anomaly combinations, calculate the transmission time (time difference from the first combination to subsequent combinations), the growth rate of the number of combinations, and the number of feature dimensions involved. For example, a trend anomaly lasts for 5 periods, has a maximum deviation of 40%, and an average growth rate of 8% per sampling period; an imbalance transmission chain takes 3 sampling periods and involves 3 types of feature combinations, all of which are used as core trend parameters.

[0104] By combining the dynamic response characteristics of load types (e.g., the imbalance trend develops more slowly under heavy loads and more quickly under light loads), trend parameters are weighted and fused. For example, for heavy loads (e.g., transformers), the weight of the number of abnormal durations is increased to 0.4; for light loads (e.g., insulators), the weight of the deviation rate of increase is increased to 0.4. The trend strength quantification value is obtained through weighted calculation, and combined with the existence or absence of anomaly transmission chains, a complete early trend characteristic description is formed, including four core dimensions: trend strength level (weak / medium / strong), the range of anomaly-involved characteristics, the trend development rate, and the imbalance transmission state.

[0105] Validate the effectiveness of early trend features: Compare the early trend features in historical imbalance fault data with the corresponding relationship of the final fault state. If the current trend feature has a similarity of more than 80% with the early features of a certain type of mild imbalance fault, the trend feature is confirmed to be effective. If the similarity is less than 30%, the time series feature matrix and anomaly identification results are re-examined to eliminate false trends caused by sensor noise and ensure that the trend feature can truly reflect the early evolution of system imbalance.

[0106] Step 73: Compare the early trend characteristics with the preset warning levels to obtain the comparison results; determine the warning level reached by the current load imbalance risk based on the comparison results; generate the corresponding load imbalance fault warning signal according to the warning level, specifically including: preset four-level warning level standards, each level corresponding to a clear early trend characteristic threshold, as follows: Level 1 (Blue Warning, Minor Risk): Trend intensity level is weak, the range of abnormal features is ≤2 types, the trend development rate is ≤5% / sampling period, and there is no imbalance transmission chain; Level 2 (Yellow Warning, General Risk): Trend intensity level is weak- Level 1 (Orange Alert, High Risk): The anomaly involves 2-3 types of features, with a trend development rate of 5%-10% per sampling period, and a short transmission chain exists (transmission time ≥ 2 sampling periods); Level 2 (Orange Alert, High Risk): The trend intensity level is medium, the anomaly involves ≥ 3 types of features, with a trend development rate of 10%-20% per sampling period, and a complete transmission chain exists (transmission time 1-2 sampling periods); Level 3 (Red Alert, Severe Risk): The trend intensity level is "strong", the anomaly involves features covering all core dimensions, with a trend development rate ≥ 20% per sampling period, a complete transmission chain, and an increasing magnitude of anomalies at each stage.

[0107] The early trend features obtained in step 72 are decomposed into four comparison indicators: trend strength, feature coverage, development rate, and transmission chain status. Each indicator corresponds to a quantitative threshold or qualitative description in the early warning level standard. For example, the trend strength indicator corresponds to weak / medium / strong levels, the feature coverage corresponds to the range of ≤2 categories / 2-3 categories / ≥3 categories, the development rate corresponds to a specific percentage threshold, and the transmission chain status corresponds to descriptions such as none / short / complete.

[0108] Each of the four indicators representing the current trend is matched against the four-level early warning criteria, and the number of successfully matched indicators at each warning level is counted. If a level has ≥3 successfully matched indicators, it is preliminarily determined to be that warning level. If two levels have the same number of successfully matched indicators (e.g., both Level 2 and Level 3 have 2 matched indicators), the core indicator priority judgment rule is activated. The core indicators are ranked by importance as follows: trend development rate > transmission chain status > feature coverage > trend strength. The level matched by the core indicators is used as the final preliminary judgment result.

[0109] Level calibration should be performed based on environmental and load conditions: If the initial assessment indicates a level of 2 or higher, further adjustments should be made based on current environmental disturbance data (such as the presence of strong gusts exceeding 15 m / s) and load status (such as whether the load is in an accelerated lifting phase). For example, if a sudden strong gust causes the trend characteristics to reach the level 2 standard, but the gust duration is less than one sampling period and the load has not entered an unstable oscillation state, the level should be downgraded to level 1; if the load is in an accelerated phase and the trend characteristics reach the level 3 standard, the level should be upgraded to level 4 because the acceleration process can easily amplify the risk of imbalance.

[0110] Final determination of warning level and recording of basis: After calibration, the final warning level of the current load imbalance risk is output, and a level determination report is generated. The report clearly records the matching status of each comparison indicator, the judgment role of the core indicators, the calibration basis of the environment and load conditions, and the comparison results with historical similar conditions, to ensure the traceability and rationality of the level determination.

[0111] A standardized structure for early warning signals is constructed, comprising six core fields: signal identifier, warning level, generation time, core anomaly information, handling recommendations, and recipients. The signal identifier uses a coding rule of warning level-timestamp-serial number (e.g., red-20251204153000-001) to ensure signal uniqueness; the warning level directly indicates the final judgment result (blue / yellow / orange / red); the generation time is accurate to milliseconds and synchronized with the sampling time of the feature vector; the core anomaly information extracts key anomalies and trend characteristics (e.g., K value = 9, tether tension fluctuation variance exceeds the standard by 3 times, an imbalance transmission chain has formed); the handling recommendations are based on pre-set corresponding operational guidelines according to the warning level; and the recipients include three core nodes: ground control terminals, various UAV flight control systems, and the emergency repair command center.

[0112] Differentiated signal content is generated based on the warning level: Level 1 (blue) warning signal mainly provides information, with concise core anomaly information. The recommended action is to closely monitor the attitude and load swing of each UAV, temporarily refrain from adjusting lifting parameters, and set the signal transmission priority to low, only displaying a pop-up notification on the ground control terminal; Level 2 (yellow) warning signal enhances the details of the anomaly, listing all trending anomalies, and recommends to slowly reduce the lifting speed, adjust the altitude of the lead UAV to balance the pull, set the signal transmission priority to medium, provide audible and visual notifications on the ground terminal, and have the UAV flight control system receive and display the warning information; Level 3 (orange) warning signal... The system includes real-time data curves, core anomaly information including anomaly evolution trend charts, and recommendations for handling the situation: "Immediately stop hoisting and acceleration, execute UAV attitude coordination adjustment commands, set signal priority to high, ensure all recipients receive the signal synchronously, and the UAV flight control system enters early warning response mode; Level 4 (red) early warning signal activates emergency push mechanism, core anomaly information highlights the warning to immediately hover, recommendations for handling the situation: all UAVs must hover immediately, release tether buffer devices, ground personnel should prepare for emergency, set signal priority to the highest, use dual-channel transmission of wireless broadcast and satellite backup to ensure no blind spots, and simultaneously trigger ground emergency alarm devices."

[0113] like Figure 2 As shown, embodiments of the present invention also propose a multi-machine hoisting load imbalance fault early warning system for mountainous power emergency operations, comprising: The data acquisition module is used to collect multi-source sensor data during multi-machine hoisting. The multi-source sensor data includes the tether tension data of each UAV, the load swing status data, the flight attitude data of each UAV, and the environmental disturbance data in the mountainous area. The conversion module is used to map and convert multi-source sensor data into a three-dimensional spatial data array, with each data sampling position corresponding to a spatial coordinate. The module is used to construct a continuous equilibrium surface representing the dynamic equilibrium state of multiple cranes based on a three-dimensional spatial data array; and to determine the vertical projection points of the center of gravity of four UAVs on the continuous equilibrium surface. The solver module is used to construct a four-boundary feature configuration region based on vertical projection points; and to solve for the geometric imbalance correction coefficient based on the intersection of the four-boundary feature configuration region and the feature boundary of the continuous equilibrium surface. The processing module is used to generate a comprehensive load imbalance feature vector based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the current power repair load type being hoisted. The early warning module is used to perform collaborative abnormal pattern recognition on the comprehensive load imbalance feature vector based on the comprehensive load imbalance feature vector, identify early trend features of load imbalance, and generate a load imbalance fault early warning signal in real time when the load imbalance risk is detected to reach the preset early warning level.

[0114] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

Claims

1. A method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas, characterized in that, The method includes: Collect multi-source sensor data during multi-machine hoisting. The multi-source sensor data includes tether tension data of each UAV, load swing status data, flight attitude data of each UAV, and mountainous environmental disturbance data. Multi-source sensor data is mapped and converted into a three-dimensional spatial data array, with each data sampling location corresponding to a spatial coordinate. Based on a three-dimensional spatial data array, a continuous equilibrium surface is constructed to characterize the dynamic equilibrium state of multiple cranes. Determine the vertical projection points of the four UAV center of gravity positions onto the continuous equilibrium surface; Construct a four-boundary feature configuration region based on the vertical projection points; solve for the geometric imbalance correction coefficient based on the intersection of the four-boundary feature configuration region and the feature boundary of the continuous equilibrium surface. Based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the current power repair load type being hoisted, a comprehensive load imbalance feature vector is generated. Based on the comprehensive load imbalance feature vector, collaborative abnormal pattern recognition is performed on the comprehensive load imbalance feature vector to identify early trend characteristics of load imbalance. When the load imbalance risk is detected to reach the preset warning level, a load imbalance fault warning signal is generated in real time.

2. The method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas according to claim 1, characterized in that, Multi-source sensor data is mapped and converted into a three-dimensional spatial data array, where each data sampling location corresponds to a spatial coordinate, including: Based on the load swing state data, determine the instantaneous center of gravity position of the load, and construct a global spatial reference system with the instantaneous center of gravity position as the origin; in the global spatial reference system, calculate the spatial position coordinates of each UAV. Based on the spatial coordinates of each UAV, the mapping coordinates of the connection point of each tether above the load in the global spatial reference system are calculated; the corresponding UAV tether tension data are associated with these mapping coordinates to obtain the first data; the load swing state data are associated with the origin coordinates of the global spatial reference system to obtain the second data; the flight attitude data of each UAV are associated with the spatial coordinates of each UAV to obtain the third data; based on the actual measurement location and direction of the mountainous environmental disturbance data, the corresponding spatial influence source coordinates are determined in the global spatial reference system, and the environmental disturbance data are associated with the spatial influence source coordinates to obtain the fourth data. The first, second, third, and fourth data are aggregated to obtain a three-dimensional spatial data slice; By combining three-dimensional spatial data slices from multiple consecutive sampling times in chronological order, a three-dimensional spatial data array is obtained.

3. The method for early warning of load imbalance faults in multi-machine hoisting in mountainous power emergency situations according to claim 2, characterized in that, Based on a three-dimensional spatial data array, a continuous equilibrium surface is constructed to characterize the dynamic equilibrium state of multiple cranes, including: Extract the three-dimensional spatial data slice corresponding to the current sampling time from the three-dimensional spatial data array, obtain the spatial position coordinates of all data sampling points in the three-dimensional spatial data slice, and mark the sensor data value associated with each coordinate point as the equilibrium state observation value of the coordinate point to obtain a set of marked spatial points; Using the coordinates of all spatial locations in the marked set of spatial points as input, perform 3D convex hull calculation to generate a minimum convex hull polyhedron that encloses all input points. The vertices of the minimum convex hull polyhedron are a subset of the set of spatial points, and its surface is composed of multiple triangular facets. Each triangular facet inherits the equilibrium state observations associated with its vertices. Based on the minimum convex hull polyhedron, topological subdivision is performed within each triangular facet, new vertices are inserted and connected to form a dense triangular mesh; the dense triangular mesh is iteratively smoothed and filtered to eliminate the sharp edges formed by the original convex hull vertices, and finally a continuous balanced surface is generated.

4. The method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas according to claim 3, characterized in that, Determine the vertical projection points of the four UAV center of gravity positions onto the continuous equilibrium surface, including: From the three-dimensional spatial data array, extract the spatial position coordinates of each UAV as the actual position of the center of gravity of each UAV in three-dimensional space; Based on the attitude data and gravitational acceleration information corresponding to the spatial position coordinates of each UAV, the vertical projection direction perpendicular to the local horizontal plane is calculated. The actual position of the center of gravity of each UAV in three-dimensional space is geometrically projected onto the continuous equilibrium surface along the vertical projection direction, and the spatial intersection point of each projection line and the continuous equilibrium surface is calculated. The spatial intersection of each projection line with the continuous equilibrium surface is recorded as the vertical projection point of the corresponding UAV's center of gravity on the continuous equilibrium surface.

5. A method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas according to claim 4, characterized in that, Construct a four-boundary feature configuration region based on vertical projection points; Based on the intersection of the four-boundary feature configuration region and the feature boundary of the continuous equilibrium surface, the geometric imbalance correction coefficient is solved, including: Based on the spatial position of the vertical projection point on the continuous equilibrium surface, and in accordance with the preset UAV spatial orientation sequence, adjacent projection points are connected in sequence to construct a spatial quadrilateral on the continuous equilibrium surface. This spatial quadrilateral is defined as the four-boundary feature configuration region. Based on the spatial distribution of all data points on the continuous equilibrium surface, the outer contour boundary line of the continuous equilibrium surface is extracted as the feature boundary of the continuous equilibrium surface. Calculate the spatial intersection points of the edges of the four-boundary feature configuration region and the feature boundary lines of the continuous equilibrium surface to obtain the feature boundary intersection points; Using the continuous equilibrium surface as the top surface and the plane passing through the four vertices of the four boundary feature configuration region and parallel to the horizontal plane as the bottom surface, calculate the spatial volume between the continuous equilibrium surface and the bottom surface, and use it as the first reference volume; Based on the intersection of the feature boundaries, a closed surface region is defined on the continuous equilibrium surface by the four-boundary feature configuration region and the surface feature boundary; the spatial volume between the closed surface region and the horizontal bottom surface is calculated as the second actual volume; By comparing and analyzing the degree of difference between the first reference volume and the second actual volume, the geometric imbalance correction coefficient is calculated.

6. A method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas, as described in claim 5, is characterized in that... Based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the current power repair load being hoisted, a comprehensive load imbalance feature vector is generated, including: Based on the geometric imbalance correction coefficient, it is used as a scalar feature characterizing the imbalance state of the four-boundary feature configuration region. Extracting key time-domain and frequency-domain features from multi-source sensor data; Based on the power repair load type specified in the current hoisting task, static and dynamic characteristic parameters corresponding to key time-domain and frequency-domain features are retrieved from the preset load characteristic database; The geometric imbalance correction coefficient, key time-domain and frequency-domain features, and static and dynamic characteristic parameters are normalized and spliced ​​according to a preset feature order to form a comprehensive load imbalance feature vector.

7. A method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas according to claim 6, characterized in that, Key time-domain and frequency-domain characteristics include: real-time values, mean values, and variances of tether tension for each UAV; angular velocity and angular acceleration of load swing; pitch and roll angles for each UAV; and instantaneous wind speed and gust intensity during mountainous environmental disturbances.

8. A method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas, as described in claim 7, is characterized in that... Static and dynamic characteristic parameters include the rated mass of the load, the position of the center of mass, the aerodynamic shape coefficient, and the maximum permissible swing angle threshold.

9. A method for early warning of load imbalance faults in multi-machine emergency power supply in mountainous areas, as described in claim 8, is characterized in that... Based on the comprehensive load imbalance feature vector, collaborative anomaly pattern recognition is performed on the comprehensive load imbalance feature vector to identify early trend characteristics of load imbalance. When the load imbalance risk is detected to reach a preset warning level, a load imbalance fault warning signal is generated in real time, including: Based on the comprehensive load imbalance feature vector, isolated anomalies and combinations of abnormal features that deviate from the normal coordination mode are identified; Based on the current moment and the combination of isolated outliers and outlier features from multiple consecutive previous moments, early trend characteristics of load imbalance are determined. The early trend characteristics are compared with the preset warning level to obtain the comparison results; based on the comparison results, the warning level reached by the current load imbalance risk is determined. Based on the warning level, a corresponding load imbalance fault warning signal is generated.

10. A mountainous area power emergency multi-machine hoisting load imbalance fault early warning system, the system implementing the method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect multi-source sensor data during multi-machine hoisting. The multi-source sensor data includes the tether tension data of each UAV, the load swing status data, the flight attitude data of each UAV, and the environmental disturbance data in the mountainous area. The conversion module is used to map and convert multi-source sensor data into a three-dimensional spatial data array, with each data sampling position corresponding to a spatial coordinate. The module is used to construct a continuous equilibrium surface that characterizes the dynamic equilibrium state of multiple cranes based on a three-dimensional spatial data array. Determine the vertical projection points of the four UAV center of gravity positions onto the continuous equilibrium surface; The solver module is used to construct a four-boundary feature configuration region based on vertical projection points; and to solve for the geometric imbalance correction coefficient based on the intersection of the four-boundary feature configuration region and the feature boundary of the continuous equilibrium surface. The processing module is used to generate a comprehensive load imbalance feature vector based on the geometric imbalance correction coefficient, multi-source sensor data, and the characteristics of the current power emergency repair load type being hoisted. The early warning module is used to perform collaborative abnormal pattern recognition on the comprehensive load imbalance feature vector based on the comprehensive load imbalance feature vector, identify early trend features of load imbalance, and generate a load imbalance fault early warning signal in real time when the load imbalance risk is detected to reach the preset early warning level.

Images