A substation low-altitude target monitoring method based on 5G-A sensing integration technology

Abstract

The application discloses a substation low-altitude target monitoring method based on 5G-A integrated sensing technology, and relates to the technical field of integrated sensing and defense, which comprises the following steps: acquiring an electronic map of a substation and key equipment position information, and constructing a three-dimensional semantic map of the substation area; combining the kinematic characteristics of a unique low-altitude target continuous trajectory and the key equipment function semantic annotation in the three-dimensional semantic map of the substation area, performing behavior pattern recognition and intention understanding on the unique low-altitude target, and outputting a target type label, an intention label and a corresponding threat level; comparing the target type label, the intention label, the corresponding threat level and the unique low-altitude target continuous trajectory with an electronic fence, generating a graded alarm event and a countermeasure suggestion, and issuing a linkage instruction to a defense device and a linkage platform to trigger corresponding disposal actions. The application realizes accurate mapping from target behavior to threat intention by extracting kinematic characteristics and function semantic annotation, and improves the cognitive and predictive ability of hidden risks.

Description

Technical Field

[0001] This invention relates to the field of integrated sensing security technology, and in particular to a method for monitoring low-altitude targets in substations based on 5G-A integrated sensing technology. Background Technology

[0002] With the rapid development of 5G-A integrated sensing technology, wireless communication networks have acquired native sensing capabilities, enabling the detection and tracking of low-altitude targets through base station signals. In the field of power infrastructure protection, this technology provides a new technical path for low-altitude safety monitoring of substations, enabling applications such as drone trajectory tracking and electronic fence early warning under wide-area coverage through the networking of sensing base stations, forming a new infrastructure architecture of communication and sensing fusion. Existing research has verified the feasibility of integrated sensing technology in target positioning, speed measurement, etc., laying a theoretical foundation for the transformation of low-altitude security from the traditional radar mode to the communication network mode.

[0003] However, existing monitoring solutions based on integrated sensing still have shortcomings. In the complex electromagnetic environment of substations, multipath effects and equipment interference can cause a sharp deterioration in the signal-to-noise ratio of sensing signals, and traditional clutter suppression algorithms are unable to effectively distinguish between real targets and reflections from fixed obstacles. At the same time, existing methods lack the ability to understand the semantics of the monitoring scene and cannot correlate target behavior with the functional risks of substation equipment, resulting in a high false alarm rate. In particular, when dealing with "low, slow and small" targets, traditional parameter estimation methods are unable to achieve accurate behavioral intent discrimination because they do not incorporate spatial semantic context. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a substation low-altitude target monitoring method based on 5G-A integrated sensing technology to solve the problems of high false alarm rate and insufficient accuracy of behavior intent recognition in low-altitude target monitoring under complex electromagnetic environments.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for monitoring low-altitude targets in substations based on 5G-A integrated sensing technology. The method includes: acquiring an electronic map of the substation and key equipment location information to construct a three-dimensional semantic map of the substation area; generating an electronic fence and a monitoring task parameter set bound to the spatial coordinates of the substation area based on a pre-set security strategy and the three-dimensional semantic map; configuring multiplexed sensing and communication resources on the 5G-A integrated sensing base station side using the monitoring task parameter set to generate a beam scanning sequence; acquiring sensing signals based on the beam scanning sequence, extracting measurement parameter sequences, and performing environmental reconstruction and clutter suppression in conjunction with the three-dimensional semantic map of the substation area to generate a purified measurement parameter sequence; and performing parameter analysis on the purified measurement parameter sequence. The system estimates and detects targets to generate a set of candidate low-altitude targets. It then performs cross-station association and deduplication on the candidate low-altitude target sets obtained from multiple substations to obtain a unique low-altitude target and its continuous trajectory. Combining the kinematic characteristics of the unique low-altitude target's continuous trajectory with the semantic annotations of key equipment functions in the substation's 3D semantic map, it performs behavioral pattern recognition and intent understanding on the unique low-altitude target, outputting target type labels, intent labels, and corresponding threat levels. Finally, it compares the target type labels, intent labels, corresponding threat levels, and the unique low-altitude target's continuous trajectory with the electronic fence to generate tiered alarm events and countermeasure suggestions. It then issues linkage commands to defense equipment and the linkage platform to trigger corresponding actions.

[0007] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A integrated sensing technology described in this invention, the steps for constructing a three-dimensional semantic map of the substation area are as follows: Obtain the electronic map and key equipment location information of the substation to obtain the original spatial dataset. Perform coordinate system unification and timestamp synchronization processing on the original spatial dataset to generate a standardized spatial dataset. Based on a standardized spatial dataset, a three-dimensional geometric structure of the station area is constructed, and functional semantic annotations are performed on key equipment in the three-dimensional geometric structure of the station area to generate a semantic annotation layer. By integrating the 3D geometric structure and semantic annotation layer of the station area, a 3D semantic map of the station area is output.

[0008] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A sensing integration technology described in this invention, the steps for generating the electronic fence and monitoring task parameter set bound to the station area spatial coordinates are as follows: Risk rule sets are extracted from pre-set security policies. Based on the risk rule sets and the three-dimensional semantic map of the station area, dynamic risk scores for each area in the three-dimensional semantic map of the station area are calculated. The dynamic risk score is compared with the risk level threshold in the risk rule set. The contour of the area with the same risk level in the 3D semantic map of the station area is delineated using the isosurface extraction algorithm to generate the electronic fence boundary. Based on the risk level of the electronic fence boundary and the resource status of the 5G-A integrated sensing base station, a set of monitoring task parameters is generated.

[0009] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A integrated sensing technology described in this invention, the steps for generating the beam scanning sequence are as follows: The scanning requirement parameters of each zone of the electronic fence are extracted from the monitoring task parameter set, and the real-time resource status of the 5G-A integrated sensing base station is monitored to generate a resource status vector. A resource allocation scheme is generated by jointly processing the scanning requirement parameters and resource state vector through a multi-objective optimization algorithm, with the goal of minimizing the perception blind spot and the probability of communication interruption. The resource allocation scheme is mapped to a beam pointing angle sequence, beamwidth, and transmit power to generate an executable beam scanning sequence.

[0010] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A sensing integration technology described in this invention, the steps for generating the purified measurement parameter sequence are as follows: The 5G-A integrated sensing base station transmits and receives signals based on the beam scanning sequence, acquires sensing signals, and extracts measurement parameter sequences. Environmental reconstruction and clutter suppression are performed on the measurement parameter sequence and the three-dimensional semantic map of the station area to generate a cleaned measurement parameter sequence.

[0011] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A integrated sensing technology described in this invention, the purified measurement parameter sequence includes angle of arrival, time of arrival, received signal strength, and Doppler frequency shift.

[0012] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A integrated sensing technology described in this invention, the steps for generating a candidate low-altitude target set are as follows: Using a set of arrival angle, arrival time, received signal strength, and Doppler frequency shift from the purified measurement parameter sequence as detection points, the three-dimensional spatial coordinates and motion velocity vectors of each detection point are calculated using the MVDR algorithm, and a spatial correlation diagram of the detection points is constructed. A graph neural network is used to perform deep learning processing on the spatial correlation graph of the detection points to obtain the target space and motion parameters. Density clustering is performed on the target space and motion parameters to generate a set of candidate low-altitude targets.

[0013] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A integrated sensing technology described in this invention, the steps for obtaining a unique low-altitude target and its continuous trajectory are as follows: Based on the synchronization clock signal of the 5G-A integrated sensing base station of each substation, the candidate low-altitude target sets of different base stations are spatiotemporally aligned to generate a spatiotemporally aligned cross-station target observation sequence. Based on the spatiotemporally aligned cross-station target observation sequence, a cross-station target association graph is constructed, the multi-dimensional association degree between each observed target is calculated, and cross-station target association matching is performed on the multi-dimensional association degree. An adaptive Kalman filter is used to fuse the trajectory of the successfully matched multi-station observation data, and to resolve the conflict of matched targets, outputting a unique low-altitude target identifier and a unique low-altitude target continuous trajectory.

[0014] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A integrated sensing technology described in this invention, the steps for outputting the target type label, intent label, and corresponding threat level are as follows: Kinematic features are extracted from the continuous trajectory of a single low-altitude target and fused with multi-source features in the semantic annotations of key equipment functions in the three-dimensional semantic map of the station area to generate a spatiotemporally related feature vector. The spatiotemporal correlation feature vectors are input into the spatiotemporal graph neural network to model behavioral patterns and generate behavioral pattern embedding vectors. The behavior pattern embedding vector is classified into multiple labels and mapped to threat levels using a classifier, and the target type label, intent label and corresponding threat level are output.

[0015] As a preferred embodiment of the substation low-altitude target monitoring method based on 5G-A integrated sensing technology described in this invention, the steps of generating graded alarm events and countermeasure suggestions, and issuing linkage commands to defense equipment and linkage platform to trigger corresponding handling actions are as follows. Based on target type labels, intent labels, threat levels, and the continuous trajectory of a unique low-altitude target, a multi-dimensional risk matching degree is calculated with the electronic fence to generate a dynamic risk matching score. Based on dynamic risk matching scores and intent labels, hierarchical alarm events and countermeasure suggestions are generated through fuzzy logic decision trees. The ultra-reliable low-latency communication link of the 5G-A integrated sensing base station in the substation will send graded alarm events and countermeasure suggestions to the defense equipment and linkage platform, triggering corresponding handling actions and providing feedback on the execution status.

[0016] The beneficial effects of this invention are as follows: by extracting kinematic features and functional semantic annotations, it achieves accurate mapping from target behavior to threat intent, thereby enhancing the ability to recognize and predict hidden risks; by deeply associating the spatiotemporal movement patterns of targets with the functional risks of substation equipment, and by using a multi-source feature fusion mechanism to analyze the intent hierarchy behind the behavior, it achieves graded and accurate alarms for low-altitude intrusion events; furthermore, by understanding intent, it predicts threat evolution trends in advance, providing a reliable basis for proactive defense decisions and enhancing the intelligence and foresight of low-altitude security protection for substations. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a substation low-altitude target monitoring method based on 5G-A integrated sensing technology.

[0019] Figure 2 A flowchart for generating the purified measurement parameter sequence.

[0020] Figure 3 This is a flowchart for outputting a unique low-altitude target identifier and a unique low-altitude target's continuous trajectory.

[0021] Figure 4 A flowchart for generating tiered alarm events and countermeasure suggestions. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4This is one embodiment of the present invention, which provides a method for monitoring low-altitude targets in substations based on 5G-A integrated sensing technology, comprising the following steps: S1. Obtain the electronic map and key equipment location information of the substation, and construct a three-dimensional semantic map of the substation area; based on the preset safety policy and the three-dimensional semantic map of the substation area, generate an electronic fence and monitoring task parameter set bound to the spatial coordinates of the substation area.

[0026] S1.1: Obtain the electronic map and key equipment location information of the substation to obtain the original spatial dataset. Perform coordinate system unification and timestamp synchronization processing on the original spatial dataset to generate a standardized spatial dataset. Specifically, by calling the digital survey results already deployed in the substation or the electronic map data stored in the operation and maintenance management platform, a raw spatial dataset containing the overall layout of the substation and the location information of key equipment is obtained. Coordinate system one processing is performed on the raw spatial dataset to convert spatial data from different sources to the same geographic coordinate system, ensuring that all spatial elements are aligned under a unified spatial reference. The raw spatial dataset that has completed coordinate system one is then time-stamped and synchronized. Based on the time stamp of each data collection time, the multi-source spatial data is aligned to the same time reference point, eliminating positional deviations or inconsistencies in status caused by differences in collection time, thus forming a standardized spatial dataset.

[0027] It should be noted that the location information of key equipment refers to the geographical coordinates of equipment in the substation that plays an important role in operational safety and functional realization, such as main transformers, circuit breakers, disconnect switches, current transformers, voltage transformers, surge arresters, grounding devices, and control and protection cabinets, on the electronic map.

[0028] S1.2: Based on the standardized spatial dataset, construct the three-dimensional geometric structure of the station area, perform functional semantic annotation on the key equipment in the three-dimensional geometric structure of the station area, and generate a semantic annotation layer; Specifically, based on a standardized spatial dataset, the spatial coordinates of the terrain, building outlines, equipment foundations, and ancillary facilities within the substation area are geometrically reconstructed to generate a three-dimensional geometric structure of the substation area that includes surface meshes and voxel structures. Based on the location information of key equipment recorded in the standardized spatial dataset, the three-dimensional entities corresponding to the main transformer, circuit breaker, disconnector, current transformer, voltage transformer, surge arrester, grounding device, and control and protection cabinet are located in the three-dimensional geometric structure of the substation area. Each three-dimensional entity is then assigned a text label corresponding to its equipment type, operating function, and electrical connection relationship to form a semantic annotation layer.

[0029] S1.3: Integrate the three-dimensional geometric structure and semantic annotation layer of the station area to output a three-dimensional semantic map of the station area.

[0030] Specifically, the three-dimensional geometric structure of the station area is spatially aligned with the semantic annotation layer, so that each text identifier in the semantic annotation layer is precisely matched with its corresponding three-dimensional entity in the three-dimensional geometric structure of the station area in terms of position and range, thereby generating a three-dimensional semantic map of the station area.

[0031] S1.4: Extract the risk rule set from the pre-set security policy. Based on the risk rule set and the station area 3D semantic map, calculate the dynamic risk score for each area in the station area 3D semantic map. The expression is: ; In the formula, Represents the spatial location points in the 3D semantic map of the station area. Dynamic risk scoring at the location; Represents any spatial location point in the 3D semantic map of the station area; This represents the scaling factor used to map dimensionless ratios to risk level thresholds ranging from 0 to 100. Indicates a point in space The number of risk rules that are satisfied; This represents the total number of risk rules contained in the risk rule set extracted from the pre-configured security policy; Represents spatial location point Euclidean distance to the nearest critical equipment; Indicates a safe reference distance.

[0032] It should be noted that the safe reference distance is a benchmark distance determined by comprehensively considering the safety protection radius of typical key equipment in the substation (such as main transformers and circuit breakers), the range of electromagnetic environment influence, and the effective range of threatening targets in historical intrusion events. It is usually taken as 10 to 30 meters.

[0033] It should be noted that the pre-configured security strategy is a multi-layered and systematic set of rules, mainly including: mandatory policies and regulations based on industry standards to define the protection baseline; specific rules at the technical implementation level, such as generating electronic fences (including no-fly zones, buffer zones, and other multi-layered spatial areas), setting target classifications (such as drones and birds), intent determination rules (such as reconnaissance, intrusion, and passing through), and threat level mapping mechanisms (low, medium, and high risk); and graded response processes, such as triggering countermeasures such as logging, sound and light deterrence, or navigation deception based on risk matching, while also supporting adaptive optimization based on real-time data to ensure the compliance, accuracy, and dynamic adaptability of substation low-altitude protection.

[0034] S1.5: Compare the dynamic risk score with the risk level threshold in the risk rule set, and use the isosurface extraction algorithm to outline the areas with the same risk level in the 3D semantic map of the station area to generate the electronic fence boundary. Specifically, the dynamic risk score is compared item by item with the risk level threshold in the risk rule set to determine the risk level to which the dynamic risk score belongs. Based on the risk level threshold corresponding to the risk level, all spatial location points with the same risk level are identified in the 3D semantic map of the station area. According to the coordinate distribution of the spatial location points in 3D space, a 3D voxel grid is constructed, and each spatial location point is mapped to the corresponding voxel subgrid, and the risk level attribute of the voxel subgrid is marked. The risk level change boundary between adjacent voxel subgrids in the voxel grid is traversed, and the isosurface extraction algorithm is used to interpolate the continuous spatial point set whose risk level is exactly equal to the risk level threshold at the voxel interface where the risk level attribute transitions from below the risk level threshold to reaching or exceeding the risk level threshold. The continuous spatial point set is connected to form a triangular patch network to complete the surface reconstruction, and finally generate a closed or semi-closed geometric contour as the boundary of the electronic fence.

[0035] It should be noted that the risk level thresholds are divided into three levels: low risk, medium risk, and high risk, according to the pre-set safety strategy. The specific setting steps are as follows: Based on the typical accident case library of substations and equipment operation and maintenance procedures, determine the severity of the consequences that may be caused by various unsafe behaviors; refer to the provisions of power industry safety standards regarding work permits, safety distances, and protection levels, combine the severity of consequences with the probability of occurrence to form a three-level risk classification framework; combine actual on-site operation and maintenance experience to assign a level to each type of risk situation, and delineate the corresponding numerical range boundaries for each level to form low risk thresholds, medium risk thresholds, and high risk thresholds; in the exemplary value range, the low risk threshold is 0 to 40, the medium risk threshold is 41 to 70, and the high risk threshold is 71 to 100; these values ​​are based on a comprehensive assessment of the frequency of historical risk events, equipment safety margins, and personnel's operational fault tolerance capabilities.

[0036] S1.6: Generate a set of monitoring task parameters based on the risk level of the electronic fence boundary and the resource status of the 5G-A integrated sensing base station.

[0037] Specifically, the risk level associated with the electronic fence boundary is obtained, which is low, medium or high risk; the current resource status of the 5G-A integrated sensing base station is obtained, including available bandwidth, number of sensing beams, signal coverage strength and concurrent task carrying capacity; based on the matching relationship between risk level and resource status, the corresponding monitoring cycle, sensing accuracy level, data reporting frequency and spatial sampling density are selected to form a monitoring task parameter set.

[0038] For example, if the boundary of the electronic fence is at a low risk level, and the 5G-A integrated sensing base station is under high load, then a long monitoring cycle, low sensing accuracy level, low data reporting frequency, and sparse spatial sampling density are used to form the monitoring task parameter set.

[0039] S2. Using the monitoring task parameter set, multiplexing resources for sensing and communication are configured on the 5G-A integrated sensing base station side to generate a beam scanning sequence; based on the beam scanning sequence, sensing signals are collected, measurement parameter sequences are extracted, and environmental reconstruction and clutter suppression are performed in conjunction with the three-dimensional semantic map of the station area to generate a purified measurement parameter sequence.

[0040] S2.1: Extract the scanning requirement parameters of each zone of the electronic fence from the monitoring task parameter set, monitor the real-time resource status of the 5G-A integrated sensing base station, and generate a resource status vector; Specifically, scanning requirement parameters for each zone of the electronic fence are extracted from the monitoring task parameter set. These parameters include the monitoring cycle, sensing accuracy level, data reporting frequency, and spatial sampling density. The risk level of each zone of the electronic fence is determined when the electronic fence boundary is generated, with each zone corresponding to a risk level. The monitoring task parameter set is generated based on this risk level and the resource status of the 5G-A integrated sensing base station. Therefore, the scanning requirement parameters corresponding to each zone can be directly read based on the binding relationship between each zone of the electronic fence and the monitoring task parameter set. At the same time, the currently available bandwidth, number of sensing beams, signal coverage strength, and concurrent task carrying capacity are obtained through the operation interface of the 5G-A integrated sensing base station and arranged in a fixed order to form a resource status vector.

[0041] S2.2: A resource allocation scheme is generated by jointly processing the scanning requirement parameters and resource state vector through a multi-objective optimization algorithm to minimize the perception blind spot and the probability of communication interruption. Specifically, the multi-objective optimization algorithm adopts the Pareto optimization method, defining the sensing blind zone as the proportion of the electronic fence area not covered by the beam of the 5G-A integrated sensing base station, and defining the communication interruption probability as the possibility of data reporting failure due to insufficient bandwidth or excessive concurrent tasks. Under the premise of meeting the minimum requirements of scanning parameters for each zone, the beam pointing, bandwidth allocation and task scheduling strategy of the 5G-A integrated sensing base station are adjusted to minimize both the sensing blind zone and the communication interruption probability. Finally, the beam resources, bandwidth share and scheduling time slots allocated to each zone are determined to form a resource allocation scheme.

[0042] With the goal of minimizing the perception blind spot and the probability of communication interruption, the objective function is expressed as follows: ; ; In the formula, Represent the objective function for the perception blind zone; This represents the set of beam pointing to all electronic fence zones; This represents the set of scheduling slots for all electronic fence zones; The objective function representing the probability of communication interruption; This represents the set of bandwidth shares for all electronic fence zones; Indicates the total area of ​​the electronic fence; This represents the reciprocal of the total area of ​​the electronic fence; Indicates the total number of electronic fence zones; This represents the electronic fence partition index; Indicates the first The area of ​​each electronic fence zone; This indicates an indicator function that takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. Indicates the first The actual spatial coverage ratio of each electronic fence zone under a given beam direction and scheduling; Indicates assignment to the first Beam direction of each electronic fence zone; Indicates assignment to the first Scheduling time slots for each electronic fence zone; This represents the coverage threshold, determined by the perception accuracy level, with a typical value of 0.95. This represents the reciprocal of the total number of electronic fence zones. Indicates the first The weight of each electronic fence zone is determined by its risk level. If an electronic fence zone is designated as a high-risk area (such as the area around the main transformer or control and protection cabinet), then... ; Indicates assignment to the first Bandwidth share of each electronic fence zone; Indicates the first Each electronic fence zone meets the minimum bandwidth required for data reporting frequency; Indicates the time slot during scheduling The number of concurrent sensing tasks; This indicates the maximum concurrent task carrying capacity of the 5G-A integrated sensing base station.

[0043] S2.3: Map the resource allocation scheme to a beam pointing angle sequence, beamwidth, and transmit power to generate an executable beam scanning sequence.

[0044] Specifically, the resource allocation scheme includes the beam resources, bandwidth share, and scheduling time slots allocated to each electronic fence zone. Based on the spatial location and geometric range of each zone in the 3D semantic map of the station area, the beam pointing angle sequence required to cover the zone is determined. The beam pointing angle sequence is a series of azimuth and elevation angle combinations arranged in chronological order. The beam width is determined according to the spatial scale and sensing accuracy level of the zone. Zones with larger spatial scales or lower sensing accuracy levels correspond to wider beam widths, while zones with smaller spatial scales or higher sensing accuracy levels correspond to narrower beam widths. The transmit power is determined according to the data reporting frequency and signal coverage strength requirements. Zones with higher data reporting frequencies or higher signal coverage strength requirements correspond to higher transmit power. The beam pointing angle sequence, beam width, and transmit power corresponding to each zone are arranged in the order of scheduling time slots to form an executable beam scanning sequence.

[0045] S2.4: Control the transmission and reception of signals of the 5G-A integrated sensing base station according to the beam scanning sequence, acquire the sensing signal, and extract the measurement parameter sequence; Specifically, the 5G-A integrated sensing base station adjusts the phase configuration of the antenna array according to the beam pointing angle sequence specified at each moment in the beam scanning sequence, and transmits sensing signals in the corresponding spatial direction with the set beamwidth and transmit power. After transmission, the 5G-A integrated sensing base station synchronously opens the receiving channel and receives the echo signals reflected or scattered back from the station area under the same beam pointing, forming sensing signals. Range-Doppler demodulation, angle estimation and channel response extraction are performed on the sensing signals to obtain the measurement parameters corresponding to each beam pointing. The measurement parameters corresponding to each scheduling time slot are arranged in chronological order to form a measurement parameter sequence.

[0046] S2.5: Perform environmental reconstruction and clutter suppression on the measurement parameter sequence and the three-dimensional semantic map of the station area to generate a cleaned measurement parameter sequence.

[0047] Specifically, the spatial location corresponding to each set of measurement parameters in the measurement parameter sequence is mapped to the coordinate system of the three-dimensional semantic map of the station area based on its distance, angle of arrival, and the location of the 5G-A integrated sensing base station, forming an observation point cloud. The observation point cloud is spatially compared with the three-dimensional entities in the three-dimensional semantic map of the station area. If an observation point is located on the surface or inside a known static device and its radial velocity is close to zero, the observation point is determined to be fixed clutter. All measurement parameters corresponding to the observation points determined to be fixed clutter are removed, and the remaining measurement parameters with non-zero radial velocity or located in non-device areas are retained to form a purified measurement parameter sequence.

[0048] S2.6: The purified measurement parameter sequence includes angle of arrival, time of arrival, received signal strength, and Doppler shift.

[0049] It should be noted that the angle of arrival refers to the incident direction when the sensing signal reaches the 5G-A integrated sensing base station antenna array after being reflected from the target, and is usually expressed as azimuth and elevation angles; Time of arrival refers to the time interval between the transmission of the sensing signal from the 5G-A sensing integrated base station and the receipt of the target reflected echo; Received signal strength refers to the power of the target reflected echo signal received by the 5G-A integrated sensing base station; Doppler shift refers to the offset of the received signal frequency relative to the transmitted signal frequency caused by the relative motion between the target and the 5G-A integrated sensing base station.

[0050] S3. Perform parameter estimation and target detection on the purified measurement parameter sequence to generate a candidate low-altitude target set; perform cross-station association and deduplication on the candidate low-altitude target sets obtained from multiple substations to obtain a unique low-altitude target and a unique low-altitude target continuous trajectory.

[0051] S3.1: Take a set of angle of arrival, time of arrival, received signal strength and Doppler frequency shift from the purified measurement parameter sequence as the detection points, calculate the three-dimensional spatial coordinates and motion velocity vector of each detection point through the MVDR algorithm, and construct the spatial correlation diagram of the detection points; Specifically, a set of arrival angles, arrival times, received signal strengths, and Doppler frequency shifts from the purified measurement parameter sequence are used as detection points. The three-dimensional spatial coordinates and motion velocity vectors of each detection point are calculated using the MVDR (Minimum Variance Distortionless Response) algorithm. After obtaining the three-dimensional spatial coordinates and motion velocity vectors of each detection point, the connection relationship between detection points is established between adjacent time frames based on the criteria of spatial proximity and velocity consistency, forming a spatial correlation diagram of detection points.

[0052] It should be noted that the spatial proximity criterion means that if the three-dimensional spatial distance between two detection points is less than or equal to a preset distance threshold (e.g., 10 meters) in adjacent time frames, then the two detection points are considered to be from the same target. The velocity consistency criterion means that if the angle between the motion velocity vectors corresponding to two detection points is less than or equal to a preset angle threshold (e.g., 30 degrees) in adjacent time frames, and the difference in velocity magnitude is less than or equal to a preset velocity difference threshold (e.g., 2 m / s), then the motion states of the two detection points are considered to be consistent.

[0053] The 3D spatial coordinates and velocity vectors of each detection point are calculated using the MVDR algorithm, expressed as follows: ; ; In the formula, This represents the x-coordinate of the detection point in the three-dimensional semantic map coordinate system of the station area; The horizontal axis represents the 5G-A integrated sensing base station; This indicates the speed at which electromagnetic waves propagate in the air; Indicates arrival time; Indicates the pitch angle; Indicates azimuth; This represents the ordinate of the detection point in the coordinate system of the three-dimensional semantic map of the station area; The vertical axis represents the 5G-A integrated sensing base station; This indicates the elevation coordinates of the detection point in the three-dimensional semantic map coordinate system of the station area; This indicates the altitude coordinates of the 5G-A integrated sensing base station; This represents the component of the velocity vector of the probe point in the horizontal direction; Indicates Doppler frequency shift; This indicates the operating carrier frequency of the 5G-A integrated sensing base station; This represents the component of the velocity vector of the probe point in the vertical direction; This represents the component of the velocity vector of the probe point in the height direction.

[0054] S3.2: A graph neural network is used to perform deep learning processing on the spatial correlation graph of the detection points to obtain the target space and motion parameters; Specifically, the spatial association graph of the detection points consists of the three-dimensional spatial coordinates, motion velocity vectors of each detection point, and connections established based on spatial proximity and velocity consistency criteria. Each detection point is a node in the graph, and the node attributes include the three-dimensional spatial coordinates and motion velocity vector of the detection point. The edges between nodes represent the association relationship that may belong to the same target in adjacent time frames. The spatial association graph of the detection points is input into a graph neural network. The graph neural network aggregates the three-dimensional spatial coordinates and motion velocity vector information of adjacent nodes in each layer through a multi-layer message passing mechanism, and updates the feature representation of the current node. After several layers of propagation, the graph neural network outputs the fused features corresponding to each connected subgraph. The fused features reflect the overall spatial position and motion state of the target represented by the subgraph. The target space and motion parameters are extracted from the fused features. The target space and motion parameters include the target's current position, motion direction, velocity magnitude, and trajectory continuity confidence.

[0055] It should be noted that the pre-training process of the graph neural network is based on labeled target trajectory samples from historical monitoring data. This historical monitoring data includes spatial correlation maps of detection points collected in the substation area, along with their corresponding real target spaces and motion parameters. The real target spaces and motion parameters are obtained using high-precision positioning equipment. The historical spatial correlation maps of detection points are used as input to the graph neural network, where the initial features of each node are the three-dimensional spatial coordinates and motion velocity vector of that detection point, and the edges are constructed using spatial proximity and velocity consistency criteria. The corresponding real target spaces and motion parameters serve as supervision labels, including the target center position and average velocity vector. The graph neural network employs a mean squared error loss function to calculate the L2 norm deviation between the network output and the supervision label. The optimizer is Adam, with an initial learning rate of 0.001, adjusted every 50 training cycles by a decay factor of 0.9. During training, the training, validation, and test sets are divided in an 8:1:1 ratio. An early stopping mechanism is used, and pre-training continues until the graph neural network's trajectory association accuracy and parameter estimation error on the validation set reach a preset stable threshold (e.g., trajectory association accuracy not less than 95% and average position estimation error not exceeding 1.5 meters), thus completing parameter solidification.

[0056] S3.3: Perform density clustering on the target space and motion parameters to generate a set of candidate low-altitude targets.

[0057] Specifically, the spatial location of each target is used as a clustering feature point. Based on the Euclidean distance between spatial locations and the angle between motion directions, a density clustering algorithm is used to identify closely clustered target groups with similar motion states in space. Based on the neighborhood range corresponding to the physical size and flight characteristics of typical low-altitude targets (such as drones and birds) in the three-dimensional semantic map of the station area, targets located in adjacent areas and with the same motion direction are divided into the same neighborhood. On the basis of neighborhood, targets that are interconnected and have a local density higher than the background noise level are grouped into the same cluster, and each cluster represents a physically possible independent low-altitude target. Sparse clusters with fewer targets than the minimum effective number of observations required for the low-altitude surveillance mission of the station area are removed. The center position and average motion parameters of the remaining clusters are used as representative attributes of the low-altitude targets corresponding to the cluster, forming a candidate low-altitude target set.

[0058] S3.4: Based on the synchronization clock signal of the 5G-A integrated sensing base station of each substation, perform spatiotemporal alignment of the candidate low-altitude target sets of different base stations to generate a spatiotemporally aligned cross-station target observation sequence; Specifically, each 5G-A integrated sensing base station is equipped with a synchronization clock signal based on the IEEE 1588 precision time protocol to ensure that the time base of all base stations is consistent. The synchronization clock signal marks the observation time of each candidate low-altitude target in each candidate low-altitude target set and unifies the observation time to the global time coordinate system. At the same time, the three-dimensional spatial coordinates of each candidate low-altitude target are transformed from the local coordinate system of their respective 5G-A integrated sensing base stations to the unified geographic coordinate system used by the three-dimensional semantic map of the station area. Under the unified time and unified spatial reference, candidate low-altitude targets from multiple substations are sorted according to the observation time, and the spatial positions of multi-station targets with time intervals less than the minimum observation period (e.g., 200 milliseconds) are compared. If the spatial distance is less than the correlation tolerance corresponding to the typical target size, it is determined that the same physical target is in different base stations and they are merged into a cross-station observation record. In this way, all cross-station observation records are integrated to form a spatiotemporally aligned cross-station target observation sequence.

[0059] For example, the typical target size for small multi-rotor drones is 0.5 to 1.2 meters in diameter, the typical target size for fixed-wing inspection drones is 1.5 to 2.5 meters in wingspan, and the typical target size for large birds (such as eagles and cranes) is 0.6 to 1.0 meters in body length. Taking into account the maximum size of the target and the positioning error, the spatial correlation tolerance is usually taken as 3 to 5 times the typical target size, that is, 2 to 6 meters.

[0060] S3.5: Based on the spatiotemporally aligned cross-station target observation sequence, construct a cross-station target association graph, calculate the multi-dimensional association degree between each observed target, and perform cross-station target association matching on the multi-dimensional association degree; Specifically, the spatiotemporally aligned cross-site target observation sequence includes candidate low-altitude targets observed by multiple 5G-A integrated sensing base stations in substations, along with their unified timestamps, three-dimensional spatial coordinates, and motion velocity vectors. Using each observed target as a node, potential association edges are established between observed targets at different base stations but with similar times, forming a cross-site target association graph. The multi-dimensional association degree is jointly determined by spatial distance, velocity vector angle, trajectory continuity confidence, and observation time difference. Spatial distance is the Euclidean distance between two targets in a unified geographic coordinate system; velocity vector angle is the angle between their motion directions; trajectory continuity confidence comes from the graph neural network output; and observation time difference is the time interval marked by the synchronization clock signal. The multi-dimensional association degree is mapped to association weights. Specifically, spatial distance, velocity vector angle, trajectory continuity confidence, and observation time difference are normalized to the interval between 0 and 1, and then combined using a product to obtain the association weights, forming a weighted cross-site target association graph. Under the constraint that each target is matched at most once, the matching scheme with the largest total association weight is selected to complete the cross-site target association matching.

[0061] S3.6: Adaptive Kalman filtering is used to fuse the trajectory of the successfully matched multi-station observation data, and conflict resolution is performed on the matched conflicting targets, outputting a unique low-altitude target identifier and a unique low-altitude target continuous trajectory.

[0062] Specifically, the successfully matched multi-station observation data includes spatiotemporally aligned observation points from multiple 5G-A integrated sensing base stations in substations. Each observation point has three-dimensional spatial coordinates and a motion velocity vector. The adaptive Kalman filter uses the initial observation point as the initial state value and updates the state estimate sequentially using the multi-station observation data at each time. During each update, the process noise covariance and observation noise covariance are dynamically adjusted based on the current observation residuals to adapt the filter gain to changes in target maneuverability. In the case of multiple candidate matches pointing to the same target in cross-station target association matching, the paths are sorted according to the cumulative association weight and trajectory smoothness of each matching path. The path with the highest weight and the smallest trajectory curvature is retained, and the remaining conflicting paths are eliminated. The unique observation sequence after conflict resolution is input into the adaptive Kalman filter to generate a smooth and continuous state estimation sequence. This sequence constitutes a unique continuous trajectory of the low-altitude target, and a globally unique low-altitude target identifier is assigned to this trajectory.

[0063] Example: If a target is matched with two trajectories at the same time during the association phase, one of which has a cumulative association weight of 0.92 and a small trajectory curvature, and the other has a weight of 0.76 and a trajectory that frequently turns back and forth, then the former is retained, and the latter is judged as a mismatch and removed, to ensure that each unique low-altitude target identifier corresponds to a physically reasonable continuous trajectory.

[0064] S4. Combining the kinematic characteristics of the continuous trajectory of the unique low-altitude target with the semantic annotation of the key equipment functions in the three-dimensional semantic map of the station area, the behavior pattern recognition and intent understanding of the unique low-altitude target are performed, and the target type label, intent label and corresponding threat level are output.

[0065] S4.1: Extract kinematic features from the continuous trajectory of the unique low-altitude target, and perform multi-source feature fusion with the semantic annotations of key equipment functions in the three-dimensional semantic map of the station area to generate a spatiotemporal related feature vector; Specifically, kinematic features are extracted from the continuous trajectory of the unique low-altitude target, including position sequence, velocity magnitude, rate of change of heading angle, and altitude fluctuation amplitude; the semantic annotations of key equipment functions in the three-dimensional semantic map of the station area include the equipment type and electrical function attributes of main transformers, circuit breakers, disconnect switches, current transformers, voltage transformers, surge arresters, grounding devices, and control and protection cabinets; the continuous trajectory of the unique low-altitude target is aligned in time to the coordinate system of the three-dimensional semantic map of the station area to determine the type, function category, and spatial distance of the target to the nearest key equipment at each time; the kinematic features are combined with the semantic annotations of key equipment functions at the corresponding time to form a joint description that includes the target's own dynamic behavior and environmental semantic context, generating a spatiotemporal correlation feature vector.

[0066] S4.2: Input the spatiotemporal correlation feature vector into the spatiotemporal graph neural network to model behavioral patterns and generate behavioral pattern embedding vectors; Specifically, the spatiotemporal correlation feature vector contains the coupled information of the kinematic features of the continuous trajectory of the unique low-altitude target and the semantic annotation of the key equipment functions in the three-dimensional semantic map of the station area, arranged in chronological order to form a node sequence; the spatiotemporal graph neural network uses the spatiotemporal correlation feature vector of each time step as a graph node, establishes edge connections based on temporal adjacency and spatial proximity, and constructs a spatiotemporal graph structure; through a multi-layer message passing mechanism, it aggregates the node features of adjacent time steps and neighboring spatial regions, and updates the node representation layer by layer; after several propagation layers, it performs pooling operation on all graph nodes to obtain a fixed-dimensional vector representing the overall behavior pattern, i.e., the behavior pattern embedding vector.

[0067] It should be noted that the pre-training process of the spatiotemporal graph neural network is based on the labeled behavior category samples in the historical low-altitude target monitoring data. The historical low-altitude target monitoring data includes the continuous trajectory of the only low-altitude target collected in multiple substations in the past, the corresponding spatiotemporal correlation feature vectors, and manually labeled behavior categories, including normal inspection, passing by, hovering reconnaissance, and approaching intrusion. The historical spatiotemporal correlation feature vectors are constructed into a spatiotemporal graph in chronological order, which is used as the input of the spatiotemporal graph neural network, and the corresponding behavior categories are used as supervision labels. The cross-entropy loss function is used to calculate the deviation between the classification probability distribution output by the network and the real label. The network aggregates local dynamic and semantic context features through multi-layer spatiotemporal message passing, and finally generates behavior pattern embedding vectors through global pooling, and connects to the fully connected classification head to output category prediction. The optimizer is Adam, with an initial learning rate of 0.0005, a batch size of 32, and the training set, validation set, and test set are divided in a 7:2:1 ratio. Pre-training continues until the behavior classification accuracy on the validation set reaches a stable level (e.g., stable above 90% for three consecutive training cycles and an F1 score of not less than 0.88), and the parameters are solidified.

[0068] S4.3: The behavior pattern embedding vector is classified into multiple labels and mapped to threat levels using a classifier, and the target type label, intent label and corresponding threat level are output.

[0069] Specifically, the behavior pattern embedding vector represents the coupling features of the motion and semantic environment of low-altitude targets; the classifier adopts a multilayer perceptron structure and has been trained using historical labeled data under the guidance of a pre-set security strategy. The historical data includes the behavior pattern embedding vector and its corresponding target type label (such as drone, bird, floating object), intent label (such as passing by, inspection, hovering reconnaissance, approaching intrusion), and threat level (low risk, medium risk, high risk); the behavior pattern embedding vector is fed into the classifier, and the classifier activates multiple output nodes at the same time, corresponding to a target type label, an intent label, and a threat level, respectively.

[0070] It should be noted that the classifier is trained end-to-end using a multi-label binary cross-entropy loss function with a learning rate of 0.001.

[0071] S5. Compare the target type label, intent label, corresponding threat level, and unique low-altitude target continuous trajectory with the electronic fence to generate graded alarm events and countermeasure suggestions, and issue linkage instructions to defense equipment and linkage platform to trigger corresponding handling actions.

[0072] S5.1: Based on target type labels, intent labels, threat levels, and the continuous trajectory of a unique low-altitude target, perform multi-dimensional risk matching calculations with the electronic fence to generate a dynamic risk matching score; ; In the formula, Indicates dynamic risk matching score; Indicates the target type matching factor; The target type label includes "drone", "birds", and "floating objects". Indicates the intent matching factor; Intent labels include “passing by,” “inspection,” “hovering reconnaissance,” and “approaching intrusion.” This represents the total length of the trajectory segment located inside the electronic fence boundary in the continuous trajectory of a unique low-altitude target; This represents the total length of the continuous trajectory of a unique low-altitude target. This represents the nonlinear sensitivity index.

[0073] It should be noted that the target type matching factor comes from the risk prior definition for different low-altitude target categories in the pre-set security strategy. The example values ​​are 0.2 for legal inspection drones, 0.8 for small commercial drones and 1.0 for unknown aircraft. The values ​​are based on the frequency and severity of the threats posed by various targets to the substation in historical security events. The intent matching factor is derived from the risk judgment rules for different behavioral intents in the pre-set security strategy. The example values ​​are: "passing by" 0.1, "inspection" 0.3, "hovering reconnaissance" 0.9 and "approaching intrusion" 1.0. The values ​​are based on historical experience and accident case statistics on the correlation between target behavior and potential threats in the power security regulations. The nonlinear sensitivity index originates from the nonlinear response requirements of the proportion of spatial intrusion in the field of risk assessment. The example value is 0.7. The basis for this value is to ensure that the score is sensitive to short-term high-risk intrusions while avoiding over-amplification of minor intrusions in long trajectories.

[0074] S5.2: Based on dynamic risk matching scores and intent labels, generate graded alarm events and countermeasure suggestions through fuzzy logic decision trees; Specifically, the dynamic risk matching score reflects the overall risk level between the target and the electronic fence, and the intent label describes the nature of the target's behavior. Both serve as input conditions for the fuzzy logic decision tree. The fuzzy logic decision tree branches and judges according to the rule nodes defined in the pre-set security policy. For example, "if the dynamic risk matching score is high and the intent label is approaching intrusion, then trigger a level 1 alarm and suggest starting navigation deception." Each leaf node corresponds to a level alarm event (such as level 1 alarm, level 2 alarm, and level 3 alarm) and corresponding countermeasure suggestions (such as sound and light deterrence, navigation deception, and log recording).

[0075] For example, when the dynamic risk matching score is 78 and the intent label is "hovering reconnaissance", the fuzzy logic decision tree determines that it falls into the "high risk - suspicious behavior" area, generates a level 2 alarm event, and outputs the countermeasure suggestion "turn on sound and light deterrence and continue to track".

[0076] S5.3: Through the ultra-reliable low-latency communication link of the 5G-A integrated sensing base station in the substation, graded alarm events and countermeasure suggestions are sent to the defense equipment and linkage platform, triggering corresponding handling actions and providing feedback on the execution status.

[0077] Specifically, the tiered alarm events and countermeasure suggestions include alarm levels and specific countermeasures; the 5G-A integrated sensing base station uses its ultra-reliable low-latency communication link to send tiered alarm events and countermeasure suggestions in a priority message format to defense devices such as sound and light deterrence devices and navigation deception devices, as well as the substation operation and maintenance linkage platform; after receiving the instructions, the defense devices initiate corresponding actions, such as activating strong light warnings or broadcasting false navigation signals, and the linkage platform simultaneously records the events and schedules manual review; after each defense device completes its action, it transmits the execution status back through the same ultra-reliable low-latency communication link, forming a closed-loop feedback.

[0078] In summary, this invention achieves precise mapping from target behavior to threat intent by extracting kinematic features and functional semantic annotations, thereby enhancing the ability to recognize and predict hidden risks; it deeply correlates the spatiotemporal movement patterns of targets with the functional risks of substation equipment, and uses a multi-source feature fusion mechanism to analyze the intent hierarchy behind the behavior, enabling graded and precise alarms for low-altitude intrusion events; furthermore, it predicts threat evolution trends in advance through intent understanding, providing a reliable basis for proactive defense decisions and enhancing the intelligence and foresight of low-altitude security protection for substations.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring low-altitude targets in substations based on 5G-A integrated sensing technology, characterized in that: include, Obtain electronic maps and key equipment location information of substations to construct a three-dimensional semantic map of the substation area; Based on the pre-set security policy and the three-dimensional semantic map of the station area, an electronic fence and a set of monitoring task parameters bound to the spatial coordinates of the station area are generated; Using the monitoring task parameter set, multiplexing resources for sensing and communication are configured on the 5G-A integrated sensing base station side to generate a beam scanning sequence. Sensing signals are collected based on the beam scanning sequence, and measurement parameter sequences are extracted. Environmental reconstruction and clutter suppression are then performed using a 3D semantic map of the station area to generate a purified measurement parameter sequence. The steps are as follows. The 5G-A integrated sensing base station transmits and receives signals based on the beam scanning sequence, acquires sensing signals, and extracts measurement parameter sequences. Environmental reconstruction and clutter suppression of the measurement parameter sequence and the three-dimensional semantic map of the station area to generate a purified measurement parameter sequence refers to mapping the spatial location corresponding to each set of measurement parameters in the measurement parameter sequence to the coordinate system of the three-dimensional semantic map of the station area based on its distance, angle of arrival and the location of the 5G-A integrated sensing base station, to form an observation point cloud; Spatial comparison is performed between the observation point cloud and the three-dimensional entities in the three-dimensional semantic map of the station area. If an observation point is located on the surface or inside a known static device and its radial velocity is close to zero, the observation point is determined to be a fixed clutter. All measurement parameters corresponding to the observation points determined to be fixed clutter are removed, and the remaining measurement parameters with non-zero radial velocity or located in non-equipment areas are retained to form a purified measurement parameter sequence. The purified measurement parameter sequence includes angle of arrival, time of arrival, received signal strength, and Doppler shift; The purified measurement parameter sequence is used for parameter estimation and target detection to generate a candidate low-altitude target set; the candidate low-altitude target sets obtained from multiple substations are cross-station associated and deduplicated to obtain a unique low-altitude target and a unique low-altitude target continuous trajectory. By combining the kinematic features of the continuous trajectory of a unique low-altitude target with the semantic annotations of key equipment functions in the three-dimensional semantic map of the station area, behavioral pattern recognition and intent understanding of the unique low-altitude target are performed, and target type label, intent label and corresponding threat level are output. The system compares the target type label, intent label, corresponding threat level, and continuous trajectory of the unique low-altitude target with the electronic fence to generate graded alarm events and countermeasure suggestions. It also issues linkage instructions to defense equipment and linkage platforms to trigger corresponding handling actions.

2. The substation low-altitude target monitoring method based on 5G-A sensing integration technology as described in claim 1, characterized in that: The steps for constructing the 3D semantic map of the station area are as follows: Obtain the electronic map and key equipment location information of the substation to obtain the original spatial dataset. Perform coordinate system unification and timestamp synchronization processing on the original spatial dataset to generate a standardized spatial dataset. Based on a standardized spatial dataset, a three-dimensional geometric structure of the station area is constructed, and functional semantic annotations are performed on key equipment in the three-dimensional geometric structure of the station area to generate a semantic annotation layer. By integrating the 3D geometric structure and semantic annotation layer of the station area, a 3D semantic map of the station area is output.

3. The substation low-altitude target monitoring method based on 5G-A sensing integration technology as described in claim 1, characterized in that: The steps for generating the electronic fence and monitoring task parameter set bound to the spatial coordinates of the station area are as follows: Risk rule sets are extracted from pre-set security policies. Based on the risk rule sets and the three-dimensional semantic map of the station area, dynamic risk scores for each area in the three-dimensional semantic map of the station area are calculated. The dynamic risk score is compared with the risk level threshold in the risk rule set. The contour of the area with the same risk level in the 3D semantic map of the station area is delineated using the isosurface extraction algorithm to generate the electronic fence boundary. Based on the risk level of the electronic fence boundary and the resource status of the 5G-A integrated sensing base station, a set of monitoring task parameters is generated.

4. The substation low-altitude target monitoring method based on 5G-A sensing integration technology as described in claim 1, characterized in that: The steps for generating the beam scanning sequence are as follows: The scanning requirement parameters of each zone of the electronic fence are extracted from the monitoring task parameter set, and the real-time resource status of the 5G-A integrated sensing base station is monitored to generate a resource status vector. A resource allocation scheme is generated by jointly processing the scanning requirement parameters and resource state vector through a multi-objective optimization algorithm, with the goal of minimizing the perception blind spot and the probability of communication interruption. The resource allocation scheme is mapped to a beam pointing angle sequence, beamwidth, and transmit power to generate an executable beam scanning sequence.

5. The substation low-altitude target monitoring method based on 5G-A sensing integration technology as described in claim 1, characterized in that: The steps for generating the candidate low-altitude target set are as follows: Using a set of arrival angle, arrival time, received signal strength, and Doppler frequency shift from the purified measurement parameter sequence as detection points, the three-dimensional spatial coordinates and motion velocity vectors of each detection point are calculated using the MVDR algorithm, and a spatial correlation diagram of the detection points is constructed. A graph neural network is used to perform deep learning processing on the spatial correlation graph of the detection points to obtain the target space and motion parameters. Density clustering is performed on the target space and motion parameters to generate a set of candidate low-altitude targets.

6. The substation low-altitude target monitoring method based on 5G-A sensing integration technology as described in claim 1, characterized in that: The steps to obtain a unique low-altitude target and its continuous trajectory are as follows: Based on the synchronization clock signal of the 5G-A integrated sensing base station of each substation, the candidate low-altitude target sets of different base stations are spatiotemporally aligned to generate a spatiotemporally aligned cross-station target observation sequence. Based on the spatiotemporally aligned cross-station target observation sequence, a cross-station target association graph is constructed, the multi-dimensional association degree between each observed target is calculated, and cross-station target association matching is performed on the multi-dimensional association degree. An adaptive Kalman filter is used to fuse the trajectory of the successfully matched multi-station observation data, and to resolve the conflict of matched targets, outputting a unique low-altitude target identifier and a unique low-altitude target continuous trajectory.

7. The substation low-altitude target monitoring method based on 5G-A sensing integration technology as described in claim 1, characterized in that: The steps for outputting the target type label, intent label, and corresponding threat level are as follows: Kinematic features are extracted from the continuous trajectory of a single low-altitude target and fused with multi-source features in the semantic annotations of key equipment functions in the three-dimensional semantic map of the station area to generate a spatiotemporally related feature vector. The spatiotemporal correlation feature vectors are input into the spatiotemporal graph neural network to model behavioral patterns and generate behavioral pattern embedding vectors. The behavior pattern embedding vector is classified into multiple labels and mapped to threat levels using a classifier, and the target type label, intent label and corresponding threat level are output.

8. The substation low-altitude target monitoring method based on 5G-A sensing integration technology as described in claim 1, characterized in that: The steps for generating tiered alarm events and countermeasure suggestions, and issuing linkage commands to defense devices and the linkage platform to trigger corresponding actions are as follows. Based on target type labels, intent labels, threat levels, and the continuous trajectory of a unique low-altitude target, a multi-dimensional risk matching degree is calculated with the electronic fence to generate a dynamic risk matching score. Based on dynamic risk matching scores and intent labels, hierarchical alarm events and countermeasure suggestions are generated through fuzzy logic decision trees. The ultra-reliable low-latency communication link of the 5G-A integrated sensing base station in the substation will send graded alarm events and countermeasure suggestions to the defense equipment and linkage platform, triggering corresponding handling actions and providing feedback on the execution status.

Images