Power transmission channel risk target alarm processing method and device based on electronic fence

Abstract

The application provides a power transmission channel risk target alarm processing method and device based on an electronic fence, relates to the technical field of intelligent monitoring and target detection of power transmission channels, and comprises the following steps: acquiring historical risk target data and extracting multi-dimensional risk features, obtaining initial clustering clusters based on weighted distance clustering, determining clusters meeting preset risk screening conditions as low-risk candidate areas, performing boundary clipping on the candidate areas based on safety space constraint conditions to generate low-risk electronic fences, acquiring a current risk target, and performing alarm suppression or release operations based on motion state prediction results in response to the target falling into the fence. The application realizes accurate compliance suppression of low-risk alarms and dynamic management and control of mobile risks, and improves operation and maintenance efficiency.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring and target detection technology for power transmission channels, specifically to a method and device for handling risk target alarms in power transmission channels based on electronic fences. Background Technology

[0002] In intelligent monitoring scenarios for power transmission channels, target detection algorithms are typically used to identify risky targets such as construction machinery and wildfires, combined with electronic fence technology for regional early warning. However, existing electronic fences are mostly generated based on static spatial rules or single-location clustering, lacking multi-dimensional assessment of target risk attributes and constraints from power facility safety rules. This can easily lead to the misclassification of high-frequency but low-risk permanent areas as alarm suppression zones, or even encroachment on legally protected areas of power lines, resulting in compliance risks. At the same time, existing static fence mechanisms cannot detect changes in the movement of targets. Moving risky targets that fall within the fence but are approaching the conductor often receive indiscriminate suppression, resulting in serious underreporting of high-risk hazards. This leads to a proliferation of invalid alarms and underreporting of real risks, hindering the improvement of operation and maintenance efficiency and safety management levels. Summary of the Invention

[0003] To address the problems of existing technologies, such as crude low-risk alarm filtering mechanisms, easy omission of moving risks by static fences, and lack of power safety regulations, this application proposes a method and device for processing alarms of risk targets in power transmission channels based on electronic fences, which achieves accurate and compliant suppression of low-risk alarms and dynamic control of moving risk targets.

[0004] To achieve the above objectives, this application adopts the following technical solution: A method for handling alarms for risk targets in power transmission channels based on electronic fences includes: acquiring historical risk target data and extracting multi-dimensional risk features from the historical risk target data; calculating the weighted distance between historical risk target data based on the weight coefficients corresponding to the multi-dimensional risk features, and clustering the historical risk target data based on the weighted distance to obtain initial clusters; determining the initial clusters that meet preset risk screening conditions as low-risk candidate regions, the preset risk screening conditions including an average risk weight less than a preset weight threshold, an occurrence frequency greater than a preset frequency threshold, and an average distance from a preset reference object greater than a preset distance threshold; performing boundary clipping on the low-risk candidate regions based on preset safety space constraints to generate a low-risk electronic fence, the boundary clipping including: generating an initial polygonal region based on the coordinates of data points within the low-risk candidate region, acquiring the constraint polygonal region corresponding to the preset safety space constraints, performing a geometric difference operation on the initial polygonal region and the constraint polygonal region to obtain the clipped polygonal region; acquiring the current risk target in the current monitoring scenario; and, in response to the current risk target falling into the low-risk electronic fence, performing an alarm suppression operation or an alarm cancellation operation based on the motion state prediction result of the current risk target.

[0005] The above solution dynamically generates accurate and compliant low-risk electronic fences by integrating multi-dimensional risk feature clustering with power safety spatial constraint pruning. It also achieves adaptive hierarchical control of alarms by combining motion state prediction, thereby effectively filtering low-risk repetitive alarms while strictly adhering to the safety bottom line and preventing the underreporting of movement risks.

[0006] Preferably, based on the motion state prediction result of the current risk target, an alarm suppression operation or an alarm cancellation operation is performed, including: acquiring the target trajectory of the current risk target in consecutive image frames; predicting the motion direction, motion speed, and size change rate of the current risk target based on the target trajectory; performing an alarm suppression operation on the current risk target in response to the motion direction moving away from a preset high-risk area, the motion speed being lower than a preset speed threshold, and the size change rate not exceeding a preset change threshold; and performing an alarm cancellation operation on the current risk target and triggering an alarm in response to the motion direction pointing to a preset high-risk area, the prediction that the target will invade the preset high-risk area within a preset number of frames in the future, or the size change rate exceeding a preset change threshold.

[0007] Preferably, the method further includes: monitoring the alarm frequency within a unit of time within a low-risk electronic fence; in response to the alarm frequency being greater than a preset multiple of the historical average for the same period, temporarily lifting the alarm suppression permission of the low-risk electronic fence, and outputting all alarm information within the low-risk electronic fence.

[0008] As a preferred embodiment, before extracting the multi-dimensional risk features of historical risk target data, the method further includes: denoising the historical risk target data and removing isolated data points that appear only once and whose continuous dwell time is less than a preset time.

[0009] Preferably, the method further includes: obtaining the geographic coordinates and camera calibration parameters of the monitoring points; establishing a conversion relationship between image pixel coordinates and geodetic coordinates based on the geographic coordinates and camera calibration parameters; and mapping the pixel coordinates of low-risk candidate areas to real geographic coordinates based on the conversion relationship, so as to achieve spatial alignment between low-risk candidate areas and power facility ledger data.

[0010] As a preferred method, the current risk target in the current monitoring scenario is obtained by: acquiring real-time collected monitoring images; inputting the monitoring images into the target detection model and outputting preliminary detection results including target category, bounding box, confidence level and center point coordinates; and determining the current risk target based on the preliminary detection results.

[0011] Preferably, after performing a geometric difference operation on the initial polygonal region and the constrained polygonal region to obtain the clipped polygonal region, the method further includes: in response to the area of ​​the clipped polygonal region satisfying a preset validity condition, determining the clipped polygonal region as a low-risk electronic fence.

[0012] Preferably, the method further includes: obtaining updated historical risk target data according to a preset update cycle or in response to triggering conditions; and re-executing the steps of extracting multi-dimensional risk features, cluster analysis, and boundary trimming based on the updated historical risk target data to update the low-risk electronic fence.

[0013] Furthermore, this application also provides a power transmission channel risk target alarm processing device based on electronic fence, used to execute the power transmission channel risk target alarm processing method based on electronic fence described in any of the above claims, including: a feature extraction module, used to acquire historical risk target data and extract multi-dimensional risk features from the historical risk target data, the multi-dimensional risk features including spatial location features, semantic risk features, temporal statistical features, and relative distance features; a clustering analysis module, used to calculate the weighted distance between historical risk target data based on the weight coefficients corresponding to each feature dimension, cluster the historical risk target data based on the weighted distance to obtain initial clusters, and determine the initial clusters that meet the preset risk screening conditions. The system includes: a low-risk candidate area; a fence generation module, used to trim the boundaries of the low-risk candidate area based on preset safety space constraints to generate a low-risk electronic fence; boundary trimming including: generating an initial polygonal region based on the coordinates of data points within the low-risk candidate area; obtaining the constraint polygonal region corresponding to the preset safety space constraints; and performing a geometric difference operation between the initial polygonal region and the constraint polygonal region to obtain the trimmed polygonal region; a target acquisition module, used to acquire the current risk target in the current monitoring scene; and an alarm control module, used to respond to the current risk target falling into the low-risk electronic fence by performing alarm suppression or alarm cancellation operations based on the predicted motion state of the current risk target.

[0014] In addition, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method for handling alarms of power transmission channel risk targets based on electronic fences.

[0015] Beneficial effects: The electronic fence-based alarm processing method for power transmission channel risk targets provided in this application extracts multi-dimensional features such as spatial location, semantic risk, temporal statistics, and relative distance of historical risk targets for weighted clustering. Combined with preset safety space constraints, candidate regions are geometrically pruned to generate dynamic low-risk electronic fences. Then, based on the current target's motion state prediction results, adaptive alarm suppression or cancellation operations are performed. This mechanism changes the traditional coarse filtering mode that relies solely on static spatial rules. It accurately defines low-risk areas from a data-driven perspective and ensures compliance with power safety regulations. Simultaneously, it introduces motion situation awareness capabilities, achieving closed-loop management of maintaining and suppressing static low-risk targets and immediately canceling dynamic high-risk targets. This significantly reduces the proportion of invalid alarms while effectively avoiding missed alarms for moving risk targets, improving the accuracy and operational efficiency of intelligent monitoring of power transmission channels.

[0016] This application establishes a refined alarm classification and control logic by acquiring the target trajectory and predicting its movement direction, speed, and size change rate. When the target exhibits static characteristics such as being far from high-risk areas, moving at low speed, and having stable dimensions, suppression is implemented. Conversely, when the target exhibits dynamic threat characteristics such as pointing towards high-risk areas, predicted intrusion, or a sudden increase in size, suppression is immediately lifted and an alarm is triggered. This dynamic discrimination mechanism based on movement trends compensates for the deficiency of static fences in sensing changes in target behavior, ensuring the security and timeliness of alarm suppression.

[0017] This application monitors the frequency of alarms within the fenced area and sets up an emergency mechanism for sudden increases in alarms. When the alarm frequency within a unit of time exceeds a preset multiple of the historical average for the same period, the suppression privilege is temporarily lifted. This adaptive fault-tolerant capability prevents the risk of batch missed alarms due to historical data lag or untimely fence updates, ensuring a safety baseline under extreme or special operating conditions.

[0018] This application further improves the generation quality and spatial compliance of low-risk electronic fences by removing short-term isolated data points before clustering and combining geographic coordinate mapping and area validity verification during fence generation. Noise reduction reduces the interference of accidental factors on clustering results, while coordinate alignment and geometric clipping ensure that fence boundaries strictly comply with power facility protection regulations and safety clearance requirements, eliminating the potential for fences to intrude into legally protected areas at the source. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of the power transmission channel risk target alarm processing method based on electronic fence according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the effect of the risk target alarm processing before the improvement in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the improved risk target alarm processing effect of an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0022] Example 1 like Figure 1As shown, this embodiment provides a method for handling alarms for risk targets in power transmission channels based on electronic fences. This method integrates historical data mining and real-time situational awareness to construct a complete closed-loop process from dynamic generation of low-risk areas to hierarchical alarm control, aiming to solve the problems of high false alarm rates and lack of safety compliance constraints associated with traditional static fences. Specifically, the method includes the following steps: Step S1: Obtain historical risk target data and extract multi-dimensional risk features from the historical risk target data. The multi-dimensional risk features include spatial location features, semantic risk features, temporal statistical features, and relative distance features.

[0023] Specifically, historical risk target data typically originates from alarm records or detection logs accumulated within a certain time window in the power transmission channel monitoring system. To accurately characterize the actual risk level of the targets, this embodiment abandons the traditional approach of relying solely on a single geographical location and instead employs multi-dimensional features for comprehensive description. Spatial location features characterize the absolute or relative coordinate distribution of the target within the monitoring scenario; semantic risk features reflect the potential threat level of the target category itself, such as the risk level difference between construction machinery and birds; temporal statistical features depict behavioral patterns such as the frequency of target appearance and dwell time, distinguishing between occasional events and persistent occurrences; and relative distance features quantify the spatial proximity between the target and critical power facilities such as conductors and towers. This multi-dimensional feature extraction provides a comprehensive and fine-grained data foundation for subsequent risk assessment, avoiding misjudgments due to missing information.

[0024] Step S2: Based on the weight coefficients corresponding to each feature dimension, calculate the weighted distance between historical risk target data, and cluster the historical risk target data based on the weighted distance to obtain the initial cluster.

[0025] In this process, the calculation of weighted distance is not a simple numerical summation, but rather a differentiated configuration of the importance of each dimension based on business needs. For example, in scenarios where safety distance is extremely sensitive, relative distance features can be given higher weights; while in scenarios focusing on specific high-risk targets, the weight of semantic risk features can be increased. Clustering analysis based on this weighted distance can aggregate data points that are similar in the multi-dimensional feature space to form initial clusters. It should be understood that the "weighting" mechanism here means that the clustering results no longer merely reflect geographical clustering, but integrate the comprehensive similarity of risk attributes, behavioral habits, and safety distance, thereby ensuring the consistency of the generated clusters in terms of risk semantics.

[0026] Step S3: Initial clusters that meet the preset risk screening conditions are identified as low-risk candidate regions. The preset risk screening conditions include average risk weight less than a preset weight threshold, frequency of occurrence greater than a preset frequency threshold, and average distance from preset reference objects greater than a preset distance threshold.

[0027] This step is essentially a secondary verification and refinement of the clustering results. Specifically, an average risk weight less than a preset weight threshold ensures that the overall threat level of targets within the selected area is low, excluding high-frequency but high-risk construction areas; an occurrence frequency greater than a preset frequency threshold ensures that the area has statistically significant stability, avoiding misclassifying isolated noise points as low-risk areas; and an average distance greater than a preset distance threshold from preset reference objects (such as conductors or towers) establishes a safety buffer zone in physical space. Only clusters that simultaneously meet the above three filtering conditions are considered to have the potential to be converted into low-risk electronic fences. This rigorous screening logic reduces the probability of mislabeling real risk areas as safe areas from the source.

[0028] Step S4: Based on the preset safety space constraints, the low-risk candidate region is clipped to generate a low-risk electronic fence. The clipping includes: generating an initial polygonal region based on the coordinates of data points within the low-risk candidate region; obtaining the constrained polygonal region corresponding to the preset safety space constraints; and performing a geometric difference operation on the initial polygonal region and the constrained polygonal region to obtain the clipped polygonal region.

[0029] In this embodiment, the definition of a low-risk electronic fence specifically refers to an alarm suppression area that has undergone security compliance verification. Since the initial polygonal region generated by clustering may cross legally protected areas or intrude into the safe clearance range of intrusive equipment, external security rules must be introduced for mandatory constraints. Specifically, the system first generates an initial polygon surrounding the data points within the candidate region; simultaneously, based on industry standards such as power facility protection regulations and voltage level safe clearance, it generates a constraint polygonal region representing a prohibited or restricted area. Specifically, the construction of the constrained polygon area is based on three levels of mandatory constraint rules: First, the prohibition rule for line protection zones, which sets the legal protection zone range according to different voltage levels. For example, 5 meters outward from the conductor edge of a 10kV line, 10 meters outward for 110kV, 15 meters outward for 220kV, and 25 meters outward for 500kV. Low-risk electronic fences are not allowed to be drawn within this range. Second, the safety clearance threshold rule, which requires that the minimum horizontal and vertical distance between the inner boundary of the low-risk fence and the conductor must be greater than the minimum safety clearance for the corresponding voltage level. Third, the prohibition rule for sensitive areas, which prohibits the setting of low-risk fences within a preset range around key sensitive areas such as tower bodies, conductor crossing points, and substation entrances and exits. Subsequently, through geometric subtraction, the parts overlapping with the constrained polygons are removed from the initial polygon. This process ensures that the final low-risk electronic fence completely avoids all illegal areas in terms of geometry, achieving rigid alignment between technical logic and industry safety standards. Even if the original clustering covers some high-risk edges, the trimming operation can accurately remove them, ensuring the compliance and security of the fence application.

[0030] Step S5: Obtain the current risk target in the current monitoring scenario; in response to the current risk target falling into a low-risk electronic fence, perform alarm suppression operation or alarm cancellation operation based on the motion state prediction result of the current risk target.

[0031] This is the core control logic of this method during real-time operation. When a new risky target is detected, the system first determines whether its spatial location is within the previously generated low-risk electronic fence. If it is not within the fence, it is handled according to the standard high-risk procedure; if it is within the fence, indiscriminate suppression is not immediately implemented, but rather a dynamic decision is made based on the motion state prediction results. For example, if the prediction results show that the target is moving away from the high-risk area and its speed is stable, alarm suppression is implemented to reduce invalid interference; conversely, if the prediction shows that the target has a tendency to move closer to the guide or its size changes drastically, even if it is within the low-risk fence, suppression will be immediately lifted and an alarm will be triggered. This dual judgment mechanism effectively overcomes the defect that static fences cannot detect changes in target behavior, significantly reducing false alarms while avoiding the security risk of missed detection of moving risks.

[0032] Through the coordinated execution of steps S1 to S5 described above, this embodiment constructs an alarm processing mechanism that can adaptively learn from historical low-risk patterns, strictly adhere to power safety regulations, and possess real-time dynamic perception capabilities. This not only improves the accuracy of alarm filtering but, more importantly, establishes an intelligent management and control system based on safety and compliance.

[0033] Example 2 Building upon Example 1, this example further details the specific implementation of alarm suppression or alarm cancellation operations based on the predicted motion state of the current risk target. Performing alarm suppression or alarm cancellation operations based on the predicted motion state of the current risk target includes: acquiring the target trajectory of the current risk target in consecutive image frames; and predicting the motion direction, speed, and size change rate of the current risk target based on the target trajectory. Specifically, the system uses a target tracking algorithm to associate the same risk target in consecutive frames, forming a motion trajectory containing temporal information. Based on this, the predicted motion direction is used to determine whether the target is moving towards or away from a preset high-risk area; the predicted motion speed is used to distinguish the target's activity level—low speed usually corresponds to a stationary or loitering state, while high speed may mean rapid passage or an urgent approach; the size change rate is a key indicator characterizing the change in the relative distance between the target and the camera. In power transmission channel monitoring scenarios, if the bounding box size of the target in the image shows a continuous increasing trend, it often physically corresponds to the target moving towards the camera, and vice versa. These three indicators together constitute the construction of the target's dynamic risk situation, compensating for the deficiency that future risks cannot be predicted based solely on instantaneous position.

[0034] In response to situations where the direction of movement is away from a preset high-risk area, the movement speed is below a preset speed threshold, and the rate of change in size does not exceed a preset change threshold, an alarm suppression operation is performed on the current risk target. This combination of judgment conditions defines the state mode of safe static or safe distance. For example, when a construction machine is detected that is located within a low-risk electronic fence, but its movement vector clearly points away from the wire, and its movement is slow with a stable or reduced image size, the system can determine that the target does not pose an immediate threat, thus maintaining the alarm suppression state and only logging the information without reporting it to the main alarm queue. This multi-dimensional joint verification mechanism effectively avoids mistaking a low-risk target that is being evacuated as a potential hazard due to a single location match, improving the accuracy of alarm filtering.

[0035] In response to a target's movement pointing towards a preset high-risk area, a prediction that it will intrude into a preset high-risk area within a preset number of frames, or a size change rate exceeding a preset threshold, the system performs an alarm cancellation operation and triggers an alarm for the current high-risk target. This is a dynamic response mechanism for potentially high-risk situations. Even if the target is currently within the geometric range of a low-risk electronic fence, as long as its movement trend shows signs of approaching a high-risk area, or its size change rate indicates that it is rapidly approaching the line, the system will immediately break the original suppression state, cancel the alarm suppression, and generate a high-level alarm message. It should be understood that the preset number of frames and the preset change threshold can be configured according to the monitoring frequency and safety tolerance of the actual scenario. The core is to establish a proactive risk warning capability, ensuring that a warning is issued in time before the target actually enters the dangerous area, thereby solving the technical problem of blind spots in traditional static fences for moving risk targets.

[0036] Furthermore, to address the risk of mass underreporting caused by sudden scene changes, this embodiment also introduces an emergency mechanism for abnormal surges. This method further includes: monitoring the alarm frequency within a unit of time within a low-risk electronic fence; and, in response to an alarm frequency exceeding a preset multiple of the historical average for the same period, temporarily lifting the alarm suppression permission of the low-risk electronic fence and outputting all alarm information within the low-risk electronic fence. In actual operation and maintenance, the environment surrounding power transmission channels may undergo non-periodic changes, such as the addition of new construction sites, sudden wildfires, or seasonal vegetation growth. These changes may cause a large number of real risk targets to emerge in areas originally marked as low-risk within a short period. If the system mechanically executes the suppression rules generated from historical data, it is highly likely to cause serious underreporting. Therefore, this embodiment calculates the alarm density within the fence in real time and compares it with the historical average for the same period. When an abnormal increase in the current frequency is detected, the circuit breaker mechanism is automatically triggered. At this time, temporarily lifting the suppression permission means that the system suspends the automatic filtering function for that area, pushing all detected targets indiscriminately to the manual review end or the upper-level platform. This design retains the intelligent noise reduction capability under normal conditions while giving the system adaptive fault tolerance under extreme or unknown operating conditions, ensuring a solid safety baseline. It should be noted that temporary removal is usually time-sensitive or event-driven. Once the abnormal frequency drops or the fence is manually confirmed and updated, the system can resume normal hierarchical control logic, rather than permanently abandoning the low-risk electronic fence.

[0037] Example 3 Building upon Example 1, this example further details the preprocessing and spatial benchmark alignment mechanism for historical risk target data. Before extracting multi-dimensional risk features from the historical risk target data, the method further includes: denoising the historical risk target data and removing isolated data points that appear only once and whose continuous dwell time is less than a preset duration. Specifically, in the actual monitoring environment of power transmission channels, target detection algorithms may produce some non-realistic or sporadic detection results due to drastic changes in lighting, shadow interference, birds flying by momentarily, or sensor noise. These data points typically exhibit discrete, irregular, and extremely short-duration characteristics in their spatiotemporal distribution and do not possess statistical significance in characterizing a stable risk pattern. If such isolated noise points are directly included in subsequent cluster analysis, it is highly likely that the clustering results will contain a large number of fragmented and invalid pseudo-clusters, or that random noise will be mistakenly aggregated to form false low-risk areas, thereby interfering with the judgment of the true risk distribution. Therefore, this example sets a time-dimensional filtering threshold to treat data points that appear only once and whose continuous dwell time is less than the threshold as invalid noise and remove them. For example, in practical implementation, this preset duration can be set to 5 minutes. This means that only when a detected target continuously exists in the frame or is tracked for more than 5 minutes is it considered a valid historical risk sample for clustering. It should be understood that this preset duration is not fixed but can be adaptively adjusted according to the dynamic nature of the monitoring scenario, the detection frequency, and the business's tolerance for noise. This denoising step effectively purifies the clustering input data, improves the purity and representativeness of the initial clusters, and lays a solid data foundation for generating high-quality, low-risk electronic fences.

[0038] To ensure that the generated low-risk electronic fence meets the safety and compliance requirements of the power industry, this embodiment also introduces a cross-spatial coordinate mapping alignment mechanism. This method further includes: acquiring the geographic coordinates and camera calibration parameters of the monitoring points; establishing a transformation relationship between image pixel coordinates and geodetic coordinates based on the geographic coordinates and camera calibration parameters; and mapping the pixel coordinates of the low-risk candidate area to the real geographic coordinates based on the transformation relationship, thereby achieving spatial alignment between the low-risk candidate area and the power facility ledger data. This step is a crucial bridge connecting the visual perception space and the geographic information space. Since the risk target location output by the target detection model is based on two-dimensional information in the image pixel coordinate system, while the constraints such as line protection zones and safety clearances stipulated in the power facility protection regulations are three-dimensional geographic entities defined in a geodetic coordinate system (such as WGS-84), the two exist in different metric spaces and cannot be directly geometrically calculated. Therefore, precise coordinate transformation is necessary to anchor the visual detection results to the real geographic space. Specifically, the system first acquires the precise GPS coordinates of the tower where the monitoring equipment is located as a geographic reference point, and simultaneously reads the camera's intrinsic and extrinsic calibration parameters. Based on these parameters, a projection transformation model from the image pixel plane to the ground geographic plane is constructed. Subsequently, this transformation model is used to convert the pixel boundary points of the low-risk candidate regions obtained from the clustering in the previous steps into their corresponding latitude and longitude coordinates. This spatial alignment operation imbues the visual clustering results, which originally only existed in the image, with real geographic attributes, enabling precise overlay and comparison with vector data such as line routes and protected area boundaries in the power facility ledger. This is not only a necessary prerequisite for subsequent geometric subtraction operations to trim illegal areas, but also a core technical support to ensure that low-risk electronic fences do not intrude into legally prohibited areas and strictly adhere to safety standards in practical applications. Without this alignment mechanism, any image-based fence generation will fail power safety compliance verification and thus lose its practical application value.

[0039] Example 4 Building upon Example 1, this example further details the method for obtaining the current risk target in the current monitoring scenario. Obtaining the current risk target in the current monitoring scenario includes: acquiring real-time collected monitoring images; inputting the monitoring images into a target detection model, outputting preliminary detection results including target category, bounding box, confidence level, and center point coordinates; and determining the current risk target based on the preliminary detection results.

[0040] Specifically, real-time acquired monitoring images are typically transmitted back by visualization monitoring devices or online monitoring terminals deployed along the power transmission line. These images constitute the raw data source for risk perception. In this embodiment, the target detection model acts as a functional perception unit, its core function being to transform unstructured video frames into structured target semantic information. It should be understood that this application does not limit the specific network architecture or training strategy of the target detection model, as long as it can handle the task of identifying multiple types of hidden dangers in power transmission scenarios. In the preliminary detection results output by the model, each field carries different downstream functions: the target category is used to distinguish risk sources of different natures such as construction machinery, wildfires, and foreign objects, and is the basis for subsequent semantic risk assessment; the bounding box not only defines the spatial range of the target in the image, but is also a key geometric basis for calculating the rate of change of size to determine whether the target is close to the line; the confidence level reflects the model's certainty about the current detection results and is used to filter out low-confidence false detection noise when determining the final current risk target; the center point coordinates are the core anchor point for performing spatial location matching. The system determines whether to initiate subsequent situation prediction and hierarchical control processes by judging whether the coordinates fall within the aforementioned low-risk electronic fence area. This standardized data structure output enables efficient decoupling and seamless integration between front-end perception and back-end decision-making logic. Furthermore, in determining current risk targets based on preliminary detection results, a secondary screening can be performed using preset confidence thresholds or category whitelists. Only targets that meet business requirements and whose detection quality is up to standard are included in the alarm processing queue, thereby further improving the system's operational efficiency and anti-interference capabilities.

[0041] Example 5 In the process of generating a low-risk electronic fence in Example 1, to ensure that the final generated fence area has practical application value, this example further introduces an area validity verification mechanism. Specifically, after performing a geometric difference operation on the initial polygonal area and the constraint polygonal area to obtain the trimmed polygonal area, the method further includes: in response to the area of ​​the trimmed polygonal area satisfying a preset validity condition, determining the trimmed polygonal area as a low-risk electronic fence. Since the constraint polygonal area corresponding to the safety space constraint condition may have a large area of ​​overlap with the initial clustering area in terms of geometric shape, the remaining usable area after the geometric difference operation becomes extremely narrow or fragmented. Although such fragmented areas mathematically meet safety compliance requirements, in actual business operations they often cannot effectively contain normal low-risk target activities due to their small coverage area, or the shape distortion leads to frequent errors in subsequent target fall-into-fence judgments, losing their statistical significance and practical value as an alarm suppression area. Therefore, this example filters the trimming results by setting a preset validity condition. For example, the preset validity condition can be set as the area of ​​the trimmed polygonal area being greater than or equal to 30% of the area of ​​the initial polygonal area. Only when the remaining area retains a sufficient proportion of the original cluster range is the area considered to still possess the integrity to represent a low-risk clustering pattern, and thus it is confirmed as a valid low-risk electronic fence. Conversely, if the remaining area is below this threshold, the candidate cluster is directly discarded, and no corresponding fence is generated. This mechanism effectively avoids invalid fences from consuming system computing resources and interfering with alarm control logic, thereby improving the overall robustness of the system.

[0042] Furthermore, considering that the environment surrounding power transmission channels is not static, factors such as construction site relocation, seasonal vegetation growth, and new infrastructure additions can all cause historical risk distribution to drift. Therefore, this embodiment also establishes a dynamic update mechanism for low-risk electronic fences. The method further includes: acquiring updated historical risk target data according to a preset update cycle or in response to triggering conditions; and re-executing the steps of extracting multi-dimensional risk features, cluster analysis, and boundary trimming based on the updated historical risk target data to update the low-risk electronic fence. Specifically, the update triggering methods include two modes: time-driven and event-driven. In time-driven mode, the system can be configured to automatically initiate a full update process weekly, monthly, or quarterly, using the latest accumulated historical alarm data to re-examine risk distribution patterns, enabling the fence range to adaptively adjust to the long-term, slow evolution of the environment. In event-driven mode, when maintenance personnel manually issue update commands, or when the system detects a significant change in alarm statistical characteristics in a specific area, a local or global fence recalculation is triggered immediately. During the update process, the system fully reuses the core processing logic described in Example 1, including multi-dimensional feature extraction, weighted distance clustering, risk screening, and safety constraint pruning, only replacing the input data with the updated historical risk target dataset. This dynamic update mechanism gives low-risk electronic fences the vitality of continuous evolution, ensuring that the alarm suppression strategy always remains highly consistent with the current actual risk situation on site. This avoids false suppression or missed alarms caused by the lag in fence hardening, thereby ensuring the accuracy and reliability of the power transmission channel intelligent monitoring system during long-term operation.

[0043] Example 6 like Figure 1 As shown, this embodiment provides a power transmission channel risk target alarm processing device based on electronic fences. This device is used to execute the power transmission channel risk target alarm processing method based on electronic fences described in the previous embodiments. It should be understood that the device in this embodiment does not refer to a specific physical hardware entity, but rather to a set of functional modules implemented by software programs on a computer device or server. The modules interact and work collaboratively with each other through data interfaces or signal buses. The device includes a feature extraction module, a cluster analysis module, a fence generation module, a target acquisition module, and an alarm control module.

[0044] The feature extraction module is used to acquire historical risk target data and extract multi-dimensional risk features from this data. These features include spatial location features, semantic risk features, temporal statistical features, and relative distance features. Specifically, the feature extraction module serves as the data preprocessing entry point for the entire device. It is responsible for reading raw data from stored historical alarm logs or detection records and transforming it into structured multi-dimensional feature vectors. The output of this module is connected to the input of the clustering analysis module, providing a standardized data foundation for subsequent risk pattern mining. Through this feature engineering process, the originally discrete and unstructured historical records can be quantitatively represented, thereby supporting refined risk analysis logic.

[0045] The clustering analysis module calculates the weighted distance between historical risk target data based on the weight coefficients corresponding to each feature dimension. It then clusters the historical risk target data based on this weighted distance to obtain initial clusters, and identifies these initial clusters that meet preset risk screening conditions as low-risk candidate regions. Specifically, the clustering analysis module receives multi-dimensional feature data from the feature extraction module and internally executes a weighted distance metric and density clustering algorithm. This module not only completes the spatial aggregation of data points but also integrates risk screening logic; only clusters that simultaneously meet multiple thresholds such as average risk weight, frequency of occurrence, and safe distance are output. This design tightly combines pure mathematical clustering with business-level risk filtering, ensuring that the output low-risk candidate regions are statistically and security-wise reliable. The output of the clustering analysis module connects to the fence generation module, transmitting the filtered candidate region point set data.

[0046] The fence generation module is used to trim the boundaries of low-risk candidate areas based on preset safety space constraints, generating low-risk electronic fences. Boundary trimming includes: generating an initial polygonal region based on the coordinates of data points within the low-risk candidate region; obtaining the constraint polygonal region corresponding to the preset safety space constraints; and performing a geometric difference operation on the initial polygonal region and the constraint polygonal region to obtain the trimmed polygonal region. Specifically, the fence generation module is the core unit for ensuring alarm suppression compliance. It receives the candidate region point set output by the clustering analysis module, first constructs the initial geometry surrounding these points; simultaneously, it loads the prohibited area polygons corresponding to constraint rules such as power facility protection zones and safety clearances from external configurations or databases. Subsequently, the module internally performs a geometric difference operation to accurately remove all non-compliant parts from the initial shape, finally outputting low-risk electronic fence coordinate data that fully complies with safety regulations. The fence data generated by this module is persistently stored or directly pushed to the alarm control module as a spatial reference for real-time alarm judgment.

[0047] The target acquisition module is used to acquire current risk targets in the current monitoring scenario. Specifically, the target acquisition module acts as a bridge between the device and the front-end sensing equipment, responsible for receiving the detection result stream output by the target detection model in real time. This module parses and standardizes the input raw detection data, extracting key information such as target category, bounding box, confidence level, and center point coordinates, and encapsulates it into a unified internal data object. The output of the target acquisition module is connected to the alarm control module, providing vivid target instances for real-time situational assessment.

[0048] The alarm control module is used to respond to a current high-risk target falling into a low-risk electronic fence. Based on the predicted motion state of the current high-risk target, it performs alarm suppression or alarm cancellation operations. Specifically, the alarm control module is the decision-making center of the entire device. It consumes target streams from the target acquisition module and fence data from the fence generation module in real time, first performing spatial location matching judgment; for targets falling into the fence, it further calls motion state prediction logic (such as trajectory tracking and trend analysis), and dynamically decides whether to maintain suppression or trigger an alarm based on the prediction results. This module integrates static spatial rules and dynamic behavior perception, realizing the "initial location screening + situation verification" dual verification mechanism described in the previous embodiment. Through the close cooperation and data flow of the above modules, the device provided in this embodiment solidifies the abstract method flow into a deployable and maintainable software product form, which not only ensures the complete implementation of the technical solution, but also provides a clear architectural boundary for subsequent independent optimization and upgrades of specific modules.

[0049] Example 7 This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the power transmission channel risk target alarm processing method based on electronic fences described in any of the foregoing embodiments. Specifically, the memory, as a non-volatile storage medium, is used to persistently store the program code and related configuration data that implement the technical solution of this application; the processor, as the core of operation and control, reads the computer program in the memory through a bus and executes the corresponding instruction sequence. When the processor runs the computer program, it instantiates the various functional modules and processing steps described in the foregoing embodiments at the hardware level, including but not limited to acquiring historical risk target data and extracting multi-dimensional risk features, performing cluster analysis based on weighted distance, performing boundary pruning under safety space constraints to generate a low-risk electronic fence, and performing alarm suppression or cancellation operations based on the current risk target's motion state prediction results. It should be understood that this application does not limit the specific hardware architecture, operating system type, or deployment form of the computer device. For example, the computer device can be a local server deployed in a monitoring center, an edge computing terminal integrated on the side of the power transmission line tower, or a cloud host providing elastic computing power. Any device capable of storing and executing the aforementioned computer programs, and able to perform necessary data interaction with the front-end image acquisition equipment and the back-end alarm platform, falls within the protection scope of this application. This hardware-software combined implementation method transforms the abstract logic in the aforementioned method embodiments into a tangible product that can be practically deployed, sold, or offered for sale, thereby providing complete legal protection and technical support for the implementation of the technical solution in different application scenarios.

[0050] Example 8 To further verify the effectiveness of the technical solution in actual business scenarios, this embodiment uses intelligent monitoring of 220kV and 500kV transmission line corridors as a specific application scenario. In this scenario, the system is deployed at the back end of the visual monitoring device along the transmission line to perform real-time processing and hierarchical management of massive alarm data.

[0051] In terms of data infrastructure, historical alarm data generated from the operation and maintenance of transmission lines in a certain region in May 2025 were selected as the analysis sample. This sample set covers two typical voltage levels: 220kV and 500kV. This data records detailed information such as the spatial location, timestamp, and category labels of potential hazards, providing a sufficient statistical basis for multi-dimensional risk feature extraction and cluster analysis. It should be understood that this data scale is only illustrative; in practical applications, the sample size can be adjusted according to the scope of the monitoring area and the accumulation of historical data, as long as the basic data density requirements of the clustering algorithm are met.

[0052] In terms of object detection model configuration, this embodiment uses YOLOv8 as the basic detector. The YOLOv8 network structure consists of four core modules: the input, the backbone, the neck network, and the detection head. The backbone uses a C2f module incorporating CSPNet concepts to enhance multi-scale feature capture capabilities; the neck network uses a PAN-FPN structure to achieve deep fusion of features at different levels; and the detection head employs a decoupled structure, handling classification and regression tasks independently. YOLOv8 adopts an anchor-free detection paradigm, abandoning the traditional anchor box mechanism, and exhibits excellent performance in detection accuracy, convergence speed, and inference efficiency. During model training, a multi-task loss function is used for joint optimization. The total loss function consists of a weighted average of the classification loss, bounding box loss, and distribution focus loss, and its expression is: ; in, For classification loss, a focus adjustment coefficient is used to suppress interference from a large number of easily classified negative samples; For the complete IoU loss, the bounding box matching degree is optimized based on the overlap area, center point distance, and aspect ratio consistency. To achieve distributed focus loss, the accuracy of bounding box prediction is improved by enabling the network to quickly focus on values ​​near the target location. , , These represent the corresponding weighting factors. The synergistic effect of these loss functions ensures the robust detection performance of the model in complex power transmission scenarios.

[0053] During the parameter configuration and fence generation phases, multi-dimensional feature-weighted clustering was performed on the aforementioned historical data. Specifically, the weight coefficients for each feature dimension were configured as follows: spatial location feature weight 0.4, semantic risk feature weight 0.3, temporal statistical feature weight 0.2, and relative distance feature weight 0.1. This weight allocation reflects the characteristics of emphasizing both spatial clustering and risk attributes in power transmission scenarios. The clustering analysis adopted the weighted DBSCAN algorithm, with key parameters set as neighborhood radius ε=100 and minimum core point number MinPts=5. For low-risk candidate clusters that passed the risk screening, the corner coordinates of all hazard detection boxes within the cluster were extracted, and the two-dimensional convex hull of the point set was calculated using Andrew's Monotone Chain algorithm to obtain the initial fence polygon. In the safety space constraint trimming stage, the power facility protection specifications were strictly enforced, with the legally protected area extending 15 meters outward from the 220kV line conductor edge and 25 meters outward from the 500kV line conductor edge being set as an absolutely prohibited constraint polygon area. By using geometric difference operations, all cluster boundaries that intrude into the above-mentioned protected area are eliminated, ensuring that the generated low-risk electronic fence fully complies with industry safety compliance requirements.

[0054] In terms of operational effectiveness verification, real-time monitoring video streams were tested based on the above configuration. Test data shows that, while maintaining effective detection of genuine high-risk targets, the method achieves a comprehensive suppression rate of over 70% for low-risk recurring alarms, significantly reducing the manual review burden on maintenance personnel. More importantly, for dynamic risk targets such as mobile construction machinery, which are easily missed by traditional static fences, the false negative rate is significantly lower than that of pure spatial fence solutions. This indicates that the dynamic control mechanism based on motion state prediction effectively compensates for the shortcomings of static rules, maintaining a safety baseline while reducing noise.

[0055] like Figure 2 , Figure 3 The image shows a qualitative comparison of the alarm handling effects of risk targets before and after the improvement. Before applying the method of this application, the monitoring screen under the traditional static fence mechanism had two typical problems: the first was a large number of invalid and redundant alarms, with most detection boxes in the screen corresponding to static targets in low-risk permanent areas. These targets repeatedly triggered alarms but did not pose an actual security threat, significantly increasing the review burden on maintenance personnel; the second was the failure to report moving risk targets, such as... Figure 3 In the left image, the two red detection boxes correspond to risky targets moving towards a preset high-risk area. Because these targets initially fall within the suppression range of a traditional static low-risk fence, their alarms are indiscriminately filtered, resulting in the failure to detect real intrusion risks in a timely manner, creating a blind spot in security monitoring. After applying the method described in this application, both of these problems are effectively improved: Firstly, the compliant low-risk electronic fence generated based on historical data clustering accurately filters out redundant alarms from static, permanent targets within the low-risk area, significantly reducing the number of invalid detection boxes in the image and greatly improving the alarm signal-to-noise ratio; secondly, a motion state prediction and verification mechanism has been added for targets falling within the fence. Figure 3 In the left figure, two moving risk targets that were missed by the traditional solution were determined by trajectory analysis to be moving towards high-risk areas and showing a tendency to intrude into the line protection zone. Therefore, the alarm suppression was automatically lifted and the alarm was triggered normally, effectively avoiding the underreporting of dynamic risks.

[0056] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for handling alarms of risk targets in power transmission channels based on electronic fences, characterized in that, include: Acquire historical risk target data and extract multi-dimensional risk characteristics from the historical risk target data; Based on the weight coefficients corresponding to the multi-dimensional risk characteristics, the weighted distance between the historical risk target data is calculated, and the historical risk target data is clustered based on the weighted distance to obtain an initial cluster. The initial clusters that meet the preset risk screening conditions are identified as low-risk candidate regions. The preset risk screening conditions include an average risk weight less than a preset weight threshold, an occurrence frequency greater than a preset frequency threshold, and an average distance from a preset reference object greater than a preset distance threshold. Based on preset safety space constraints, the low-risk candidate region is clipped to generate a low-risk electronic fence. The boundary clipping includes: generating an initial polygonal region based on the data point coordinates within the low-risk candidate region; obtaining the constrained polygonal region corresponding to the preset safety space constraints; and performing a geometric difference operation between the initial polygonal region and the constrained polygonal region to obtain the clipped polygonal region. Obtain the current risk target in the current monitoring scenario; in response to the current risk target falling into the low-risk electronic fence, perform an alarm suppression operation or an alarm cancellation operation based on the motion state prediction result of the current risk target.

2. The method for handling alarms of risk targets in power transmission channels based on electronic fences according to claim 1, characterized in that, The step of performing alarm suppression or alarm cancellation operations based on the motion state prediction results of the current risk target includes: Acquire the target trajectory of the current risk target in consecutive image frames; predict the motion direction, motion speed, and size change rate of the current risk target based on the target trajectory; In response to the movement direction moving away from the preset high-risk area, the movement speed being lower than the preset speed threshold, and the size change rate not exceeding the preset change threshold, the alarm suppression operation is performed on the current risk target; In response to the movement direction pointing to the preset high-risk area, the prediction that the movement will invade the preset high-risk area within a preset number of frames, or the size change rate exceeding the preset change threshold, the alarm cancellation operation is performed on the current risk target and an alarm is triggered.

3. The method for handling alarms of risk targets in power transmission channels based on electronic fences according to claim 2, characterized in that, The method further includes: Monitor the alarm frequency per unit time within the low-risk electronic fence; In response to the alarm frequency being greater than a preset multiple of the historical average for the same period, the alarm suppression permission of the low-risk electronic fence is temporarily lifted, and all alarm information within the low-risk electronic fence is output.

4. The method for handling alarms of risk targets in power transmission channels based on electronic fences according to claim 1, characterized in that, Before extracting the multi-dimensional risk features from the historical risk target data, the method further includes: The historical risk target data is denoised to remove isolated data points that appear only once and whose continuous dwell time is less than a preset time.

5. The method for handling alarms of risk targets in power transmission channels based on electronic fences according to claim 1, characterized in that, The method further includes: Obtain the geographic coordinates and camera calibration parameters of the monitoring points; Based on the geographic coordinates and the camera calibration parameters, a conversion relationship between image pixel coordinates and geodetic coordinates is established; Based on the transformation relationship, the pixel coordinates of the low-risk candidate areas are mapped to real geographic coordinates to achieve spatial alignment between the low-risk candidate areas and the power facility ledger data.

6. The method for handling alarms of risk targets in power transmission channels based on electronic fences according to claim 1, characterized in that, The process of obtaining the current risk target in the current monitoring scenario includes: Acquire real-time monitoring images; The monitoring image is input into the target detection model, and the model outputs preliminary detection results including target category, bounding box, confidence score and center point coordinates. The current risk target is determined based on the preliminary detection results.

7. The method for handling alarms of risk targets in power transmission channels based on electronic fences according to claim 1, characterized in that, After performing a geometric difference operation on the initial polygonal region and the constrained polygonal region to obtain the clipped polygonal region, the method further includes: In response to the fact that the area of ​​the cropped polygonal region meets the preset validity condition, the cropped polygonal region is determined as the low-risk electronic fence.

8. The method for handling alarms of risk targets in power transmission channels based on electronic fences according to claim 1, characterized in that, The method further includes: According to the preset update cycle or in response to triggering conditions, obtain the updated historical risk target data; Based on the updated historical risk target data, the steps of extracting multi-dimensional risk features, cluster analysis, and boundary clipping are re-executed to update the low-risk electronic fence.

9. A power transmission channel risk target alarm processing device based on electronic fence, used to execute the power transmission channel risk target alarm processing method based on electronic fence as described in any one of claims 1 to 8, characterized in that, include: The feature extraction module is used to acquire historical risk target data and extract multi-dimensional risk features from the historical risk target data. The multi-dimensional risk features include spatial location features, semantic risk features, temporal statistical features, and relative distance features. The clustering analysis module is used to calculate the weighted distance between the historical risk target data based on the weight coefficients corresponding to each feature dimension, cluster the historical risk target data based on the weighted distance to obtain initial clusters, and determine the initial clusters that meet the preset risk screening conditions as low-risk candidate regions. The fence generation module is used to perform boundary clipping on the low-risk candidate area based on preset safety space constraints to generate a low-risk electronic fence. The boundary clipping includes: generating an initial polygonal area based on the data point coordinates within the low-risk candidate area; obtaining the constraint polygonal area corresponding to the preset safety space constraints; and performing a geometric difference operation on the initial polygonal area and the constraint polygonal area to obtain the clipped polygonal area. The target acquisition module is used to acquire the current risk targets in the current monitoring scenario; The alarm control module is used to respond to the current risk target falling into the low-risk electronic fence, and to perform alarm suppression operation or alarm cancellation operation based on the motion state prediction result of the current risk target.

10. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the power transmission channel risk target alarm processing method based on any one of claims 1 to 8.

Images