A bird-related fault identification and early warning method based on transmission tower clearance model

Abstract

The present invention discloses a bird-related fault identification and early warning method based on a transmission tower gap model, which relates to the technical field of power system safety monitoring. The method first constructs a three-dimensional transmission tower gap model containing local insulating armor, and obtains an electric field distortion feature library under various working conditions through finite element simulation. The interference behavior of bird bodies, bird nests and excrement in the safety gap is further simulated, the discharge probability is calculated and the bird images are synthesized and labeled for training the risk decision model. Combining the target detection algorithm with the integrated attention mechanism and the SPPCSPC module, the spatial coordinates of the bird target are extracted in real time and mapped to the high-risk area, and the LSTM network is used to output the discharge risk score and breakdown trend index. When the risk index exceeds the dynamically adjusted early warning threshold, an early warning signal and fault coordinates are generated. The present invention improves the accuracy and foresight of bird-related fault identification and has good engineering adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of power system safety monitoring, and in particular to a bird-related fault identification and early warning method based on a transmission tower gap model. Background Art

[0002] With the widespread promotion of cross-regional deployment of high-voltage transmission lines, the spatial intersection between bird activities and transmission equipment has become more frequent, resulting in an increase in the frequency of bird-related faults caused by bird bodies, nesting materials and excrement, which has become a safety hazard that cannot be ignored in the operation of the power system.

[0003] Traditional bird-related fault identification methods primarily rely on visible light monitoring or manual inspections, supplemented by simple image detection and static thresholding strategies. However, these methods suffer from recognition lag, a single risk modeling dimension, and unobservable electrical evolution processes. Especially in environments with varying humidity, obstructed light, or complex structures, traditional methods struggle to accurately locate bird targets or assess the probability of them triggering discharge faults. Summary of the Invention

[0004] To address the above issues, the present invention proposes a bird-related fault identification and early warning method based on a transmission tower gap model, which integrates tower electric field distribution simulation, bird-related behavior modeling and deep identification mechanism to achieve multi-dimensional modeling, real-time prediction and dynamic early warning of bird-related fault risks.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] A method for identifying and warning bird-related faults based on a transmission tower clearance model, the method comprising:

[0007] Construct a three-dimensional transmission tower gap model with partial insulation armor, calculate the electric field distribution of the three-dimensional transmission tower gap model under no-load, dry bird-wading conditions, and wet bird-wading conditions through finite element simulation, and output an electric field distortion feature library, including electric field intensity, gradient, and high-risk coordinates;

[0008] Based on the electric field distortion feature library, the spatial interference of bird bodies, nests, and excrement within a safe gap threshold is simulated, and the corresponding discharge probability and electric field distortion parameters are output. A generative adversarial network is then used to synthesize bird images annotated with category, simulated coordinates, lighting, weather, and posture information, where the lighting, weather, and posture information are used to control the synthesis of the bird images.

[0009] Associating the discharge probability, electric field distortion parameters, and bird images according to simulated coordinates to form a training data set containing the conductivity of bird nest materials and the velocity of droppings falling, which is used to train a risk decision model;

[0010] Obtain tower monitoring video streams and input them into the target detection model that integrates the global attention mechanism and the SPPCSPC module for target recognition, outputting target categories, real-time spatial coordinates, and confidence levels;

[0011] Prioritizing the mapping of the real-time spatial coordinates to the high-risk coordinate area of ​​the electric field distortion feature library, extracting the corresponding electric field strength and gradient, and combining them with the target category, confidence level, the conductivity of the bird's nest material, and the falling speed of the excrement, inputting them into the trained risk decision model. When the real-time spatial coordinates enter the safe gap threshold, the discharge risk score and breakdown trend indicator are calculated through the LSTM network;

[0012] When the discharge risk score or breakdown trend index exceeds a dynamically adjusted warning threshold, a warning signal and fault coordinates are output.

[0013] Preferably, the method also includes a fault-tolerant mechanism, specifically: when there is a missing video stream of the tower monitoring, millimeter-wave radar data is obtained, and the radar space coordinates are extracted to replace the real-time space coordinates; when the electric field distortion parameters corresponding to the real-time space coordinates are missing in the electric field distortion feature library, the training data set is called to perform interpolation compensation for the discharge probability.

[0014] Preferably, the constructing of a three-dimensional transmission tower gap model with local insulating armor specifically includes:

[0015] Input the geometric parameters of the tower body, crossarms, insulators, conductor installation positions, and insulation armor structure to generate a 3D structural model including a complete space frame.

[0016] Input the electrical parameter set and the physical parameters of the materials of each structural component, establish a material property combination, and bind the material property combination to the unit ID of the space frame one by one;

[0017] Constructing an insulating armor subunit nested in the tower body and the crossarm through a node coupling algorithm, and incorporating the material property combination;

[0018] Call the predefined transmission tower structure template library and complete automatic instantiation and combination based on structural parameters;

[0019] Grid cells are divided in the gap area between the conductor and the conductive structure, and the spatial frame, electrical parameter set and material property combination are assigned to each grid cell to construct a three-dimensional transmission tower gap model with local insulating armor for simulation.

[0020] Preferably, the electric field distribution of the three-dimensional transmission tower gap model under no-load, dry bird-wading conditions, and wet bird-wading conditions is calculated by finite element simulation, and the method includes:

[0021] Based on the divided grid cells, the gap area between the wire and the conductive structure is reconstructed into a hybrid network composed of tetrahedral cells and hexahedral cells. When the local electric field gradient change rate exceeds the preset curvature threshold, adaptive grid refinement is triggered.

[0022] The electrostatic field solver is used for no-load, dry wading bird, and wet wading bird conditions to calculate the node potential distribution and output the electric field strength and gradient of each node;

[0023] If the node meets the condition that the electric field strength is greater than the electric field strength threshold or the electric field gradient is greater than the electric field gradient critical value, the node is marked as a pre-selected high-risk coordinate;

[0024] For each pre-selected high-risk coordinate, calculate the risk sensitivity weight , the calculation formula is:

[0025] ;

[0026] Where, is the potential difference between adjacent nodes; is the maximum operating voltage of the system; is the electric field gradient modulus; is the critical breakdown field strength of air; is the safety gap threshold; is the Euclidean distance from the node to the nearest conductor; is the insulation armor protection factor, where , is the dielectric constant of the insulating armor material; 、 、 、 is the weight coefficient;

[0027] When the risk sensitivity weight When , mark the pre-selected high-risk coordinate as the final high-risk coordinate;

[0028] The electric field intensity, electric field gradient and final high-risk coordinates of all nodes are stored to form an electric field distortion feature library.

[0029] Preferably, the method of simulating the spatial interference of bird bodies, bird nests, and excrement within a safe gap threshold based on the electric field distortion feature library and outputting corresponding discharge probability and electric field distortion parameters includes:

[0030] Extracting grid cells where the gap distance between the conductor and the conductive structure is less than a set safety gap threshold from the electric field distortion feature library, constructing an interference injection area set, and arranging a simulation injection point set in the interference injection area set according to multiple sets of discrete spatial coordinates;

[0031] For each injection point in the simulation injection point set, three types of foreign body models, bird body, bird nest, and excrement, are inserted respectively. The bird body foreign body model is a long ellipsoid geometry, and the conductivity parameters are selected according to the dry wading bird working condition and the wet wading bird working condition respectively. The bird nest foreign body model adopts a polyhedron shell structure, and the conductivity parameters are derived from the predefined material property library. The excrement foreign body model is a hemispherical droplet structure, and the conductivity is ,in is the ambient humidity, is the reference conductivity;

[0032] After embedding the three types of foreign body models into the three-dimensional transmission tower gap model according to the injection point coordinates, the electrostatic field solver is called to re-solve the potential distribution after interference, output the electric field strength and gradient, and calculate the corresponding discharge probability in combination with the air breakdown strength parameter. , the formula is: ;in, is the modulus of electric field intensity; is the critical breakdown field strength of air; is the slope adjustment parameter.

[0033] Preferably, the method of synthesizing bird images and forming a training dataset includes:

[0034] A conditional generative adversarial network (CGN) is used to synthesize bird images. The generation conditions include the bird's category, simulated coordinates, light level, weather type, and posture information, including wing opening and closing and head and neck angle.

[0035] After image synthesis, the actual center coordinates of the bird target in the generated image are extracted and compared with the corresponding input simulated coordinates based on the Euclidean distance calculation method. When the spatial offset is less than the set tolerance range, the image-label consistency is confirmed.

[0036] After the image-simulation consistency judgment is completed, the synthesized bird image is bound to the corresponding discharge probability, electric field intensity, and electric field gradient value through simulation coordinates to construct associated data entries that correspond one-to-one between the image and the physical disturbance characteristics, which are used to generate a structured training dataset.

[0037] Preferably, the global attention mechanism includes a channel attention submodule and a spatial attention submodule;

[0038] The channel attention submodule generates channel attention weights by performing global average pooling and maximum pooling operations on the input feature map, and performs channel-by-channel product fusion on the channel attention weights and the original feature map; the spatial attention submodule receives an occlusion area mask set based on the tower structure configuration, and the occlusion area mask is used to adjust the spatial response of the tower body, crossarm and insulator interference area in the spatial attention calculation;

[0039] The SPPCSPC module is nested in the feature extraction path between the middle of the backbone network and the output feature layer, and includes multiple parallel convolution branches with different convolution kernel sizes and a set of channel fusion layers for connecting context features;

[0040] The target detection model detects consecutive image frames in the tower monitoring video stream frame by frame, stores candidate target outputs in a fixed time window, compares the spatial positions of the same target in consecutive frames and calculates the deviation of the center point coordinates in the frame. When the set threshold conditions are met, a majority frame voting judgment is performed to determine the final output target category, center coordinates in the image frame and confidence level; the center coordinates in the image frame are mapped into three-dimensional real-time spatial coordinates after camera parameter transformation.

[0041] Preferably, the real-time spatial coordinates are preferentially mapped to the high-risk coordinate area of ​​the electric field distortion feature library, and the corresponding electric field strength and gradient are extracted. The method includes:

[0042] A spatial index structure is constructed for high-risk coordinates in the electric field distortion feature library. The spatial index structure is constructed based on the three-dimensional Euclidean distance calculation with a radius of The spherical search domain;

[0043] The real-time spatial coordinates are input into the spatial index structure, and a spatial matching operation is performed. When the distance between any high-risk coordinate and the real-time spatial coordinate is less than or equal to When , the electric field intensity and gradient corresponding to the high-risk coordinate are extracted;

[0044] If the real-time space coordinate does not fall within the tolerance search domain of any high-risk coordinate, a global minimum distance search is performed between the real-time space coordinate and all coordinates in the electric field distortion feature library to obtain the electric field strength and gradient corresponding to the nearest neighbor coordinate, and the data sample is marked as a "non-high-risk associated sample" for model-assisted input.

[0045] Preferably, the risk decision model constructs a fusion input vector based on the training data set, and divides the fusion input vector into a static input field and a time series input field, wherein the static input field includes target category, confidence, and conductivity of the bird's nest material, and the time series input field includes real-time spatial coordinates, excrement landing speed, electric field strength and gradient;

[0046] The risk decision model includes a static feature processing branch and a time series processing branch, wherein the static feature processing branch is a subnetwork containing at least one fully connected layer, which is used to extract static feature expressions from static input fields; the time series processing branch is an LSTM network, which is used to process time series input fields and output time series feature expressions;

[0047] The static feature expression and the time series feature expression are spliced ​​and fused, and two output values ​​are generated through a fully connected layer, which are a discharge risk score and a breakdown trend indicator.

[0048] Preferably, the method for dynamically adjusting the warning threshold is:

[0049] Setting the initial discharge risk score threshold with the Breakdown Trend Indicator Threshold ;

[0050] During the operation of the system, the target category distribution, real-time spatial coordinate distribution density and historical high-risk sample triggering times within the unit time window are counted, and the category risk factor is calculated based on the high-risk bird category with the highest current occurrence frequency. , calculate the spatial interference density factor based on the frequency change of spatial coordinates falling within the safety gap threshold , calculate the historical trigger rate factor based on the number of actual discharge events or high confidence predictions within a preset time period in the past ;

[0051] The dynamically adjusted warning threshold is calculated using the weighted update formula:

[0052] ;

[0053] ;

[0054] Where, is the dynamic adjustment coefficient, with a value range of [0,1], used to control the adjustment rate; 、 They are time, The discharge risk score threshold at the moment; 、 They are time, Breakdown trend indicator threshold at the moment.

[0055] The beneficial effects of the present invention are as follows: by constructing a three-dimensional transmission tower gap model with local insulating armor, combining the finite element simulation method to systematically analyze the electric field distribution characteristics under multiple working conditions, establishing a high-risk spatial coordinate and electric field distortion feature library, and realizing the identification of sensitive areas where bird behavior may trigger discharge events. By introducing a simulated foreign body injection and discharge probability modeling mechanism, the interference effect of bird bodies, nests and excrement on the gap electric field is effectively characterized, and combined with the conditional generative adversarial network to generate bird images with rich semantic information, a multimodal training data set is constructed, thereby improving the accuracy and generalization ability of the target detection algorithm. In the target detection stage, the fusion of channel attention and spatial attention mechanisms significantly improves the recognition ability of bird targets in complex scenes; the introduction of an LSTM-based risk decision model realizes the temporal perception and trend output of spatial position, electric field parameters and conductivity changes; and by constructing a dynamically adjusted early warning threshold mechanism, the real-time and accuracy of bird-related fault identification are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort. Among them:

[0057] Figure 1 is a flow chart of the method of the present invention;

[0058] Figure 2 A flow chart of a method for constructing a three-dimensional transmission tower gap model in an embodiment of the present invention;

[0059] Figure 3 Flowchart of a method for matching real-time spatial coordinates with high-risk coordinates in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0061] like Figure 1 FIG. 1 is an embodiment of the present invention, which provides a method for identifying and warning bird-related faults based on a transmission tower clearance model, comprising the following steps:

[0062] S1: Construct a three-dimensional transmission tower gap model with local insulation armor. Calculate the electric field distribution of the three-dimensional transmission tower gap model under no-load, dry bird-wading conditions, and wet bird-wading conditions through finite element simulation. Output the electric field distortion feature library, including electric field strength, gradient, and high-risk coordinates.

[0063] A three-dimensional transmission tower gap model with local insulation armor is constructed to provide physical structure foundation and boundary input support for electric field simulation and subsequent risk assessment. Figure 2 As shown, in a preferred embodiment of the present invention, the construction steps are:

[0064] S11: Input the geometric parameters of the tower body, crossarms, insulators, conductor installation positions, and insulation armor structure to generate a three-dimensional structural model including a complete space frame;

[0065] Furthermore, the geometric parameters of the tower body refer to structural parameters, including tower type (e.g., tension tower, straight tower), tower height, cross-sectional form of each tower segment (e.g., angle steel, steel pipe cross-section), foundation coordinates, and tower body positioning information relative to the terrain. These parameters are retrieved through on-site measurements or a dispatch database and imported into the structural modeling environment as vector parameters.

[0066] Crossarm geometry refers to the installation parameters of the crossarm assembly, including installation height, connection location, crossarm span, cross-sectional shape, and orientation angle. As the core support unit for suspending conductors and insulators, the crossarm structure should be precisely aligned with the three-dimensional structural model of the tower body to form a closable three-dimensional spatial framework.

[0067] Build an insulator chain structure at the crossarm attachment points. Specific inputs include insulator string type (e.g., composite, glass, ceramic), string length, number of strings in series, material, electrical performance rating, and inclination angle. This structure hangs vertically in the direction of gravity and serves as a crucial interstitial medium for the conductor connection path. After the spatial structure framework is constructed, input conductor path information based on the lower end nodes of the insulators.

[0068] The conductor modeling uses the catenary curve approximation to represent the conductor. Combined with parameters such as phase number, minimum clearance between the conductor and the ground, height difference, phase spacing, tension boundary value, etc., the starting and ending points, sag and arc length of the conductor in the three-dimensional coordinate system are determined to generate the three-phase conductor path.

[0069] S12: Input the electrical parameter set and the physical parameters of the materials of each structural component, establish a material property combination, and bind the material property combination to the unit ID of the space frame one by one;

[0070] A set of electrical parameters, including the rated voltage of each phase conductor, potential boundary conditions, air breakdown field strength, safety clearance threshold, and insulator creepage distance, is established and prepared as the boundary input for the electric field simulation. To achieve physical consistency and material response simulation, the material properties of each structural component are further input, including dielectric constant, volume conductivity, thermal conductivity, electrical contact resistance, and surface coating resistivity. These parameters constitute a material property combination and are bound to the aforementioned spatial framework cells. Material property mapping is achieved through cell IDs to ensure accurate material property distribution in each region of the subsequent simulation.

[0071] S13: Using the node coupling algorithm, the insulating armor subunits nested in the tower body and crossarms are constructed and incorporated into the material property combination.

[0072] In the modeling of the insulation armor area, a node coupling algorithm is used to construct an insulation armor sub-unit based on the on-site armor construction drawings and parameter inputs. Inputs for this sub-unit include the armor installation start and end range (coordinate index), geometric envelope thickness, and electrical performance indicators (such as surface resistivity and dielectric strength). This sub-unit is then nested within the tower body or crossarm external node area through topological binding relationships and associated with material properties to ensure that the insulation reinforcement area has independent properties in the physical modeling.

[0073] S14: Call the predefined transmission tower structure template library to automatically complete the combination instantiation of various structural parameters, improve modeling efficiency and maintain structural consistency.

[0074] S15: Divide the gap area between the conductor and the conductive structure into grid cells, and assign the spatial frame, electrical parameter set, and material property combination to each grid cell to construct a three-dimensional transmission tower gap model with local insulation armor for simulation.

[0075] A three-dimensional finite element mesh is generated in the air gap between the conductor and any conductive structure, preferably using an irregular tetrahedron automatic meshing strategy. The aforementioned spatial framework, electrical parameter set, and material property combinations are assigned to mesh cells, achieving a consistent mapping of the physical field, boundary field, and material field, completing the construction of a three-dimensional transmission tower gap model. This model serves as the input for subsequent electric field simulation modules to analyze the electric field distribution response under different bird interference conditions.

[0076] In one embodiment of the present invention, a three-dimensional transmission tower clearance model including a conductor-insulation-ground path is constructed based on the input geometric parameters of the tower body, crossarms, insulators, conductor installation locations, and local insulation armor structures. Electrical parameters such as conductor voltage, insulator material dielectric constant, and armor material conductivity are input. Finite element simulation is then used to calculate the electric field distribution of the three-dimensional transmission tower clearance model under no-load, dry bird-wading conditions, and wet bird-wading conditions. The specific operations are as follows:

[0077] S16: Set three typical simulation conditions:

[0078] No-load state: A standard electric field distribution is established without foreign object interference as a comparison benchmark to evaluate the degree of electric field distortion when foreign objects are present. For example, a static power frequency voltage boundary (such as ±110kV or ±220kV) is applied to the conductor surface. The tower body, crossarms, and grounding metal parts are all set to zero potential. The gap space is filled only with air, with a dielectric constant of 1 and no pollution. Insulator surface contamination or climate factors are not considered. The simulation domain uses a conventional initial mesh without local refinement.

[0079] Dry wading bird state: simulates the local electric field disturbance caused by bird foreign matter (such as dry bird body, nesting material) staying in the air or approaching the wire, representing a typical sunny day bird damage scenario; for example, add an equivalent bird body geometric structure (ellipsoid, rod or wingspan model can be used) near the wire; the conductivity of the bird body model is set to (Typical values ​​range from 0.01 to 0.1 S / m), representing the conductivity of dry feather tissue. A typical approach distance of 0.1–0.3 m was set between the bird and the conductor to simulate flight approaching or resting on the crossarm. The air region maintained a dielectric constant of 1, with no contamination layer. Contamination was not considered for the insulators, which remained clean. The conductor voltage boundary was consistent with the no-load state. A local mesh refinement strategy was enabled in the simulation to refine the region between the foreign object and the conductor to capture local electric field anomalies. The results reflect the extent of the impact of dry foreign objects on the discharge path and are primarily used to identify hidden risks during normal flight.

[0080] Wet wading bird state: simulates the situation in which the surface conductivity of birds or nesting materials increases due to rain, snow, and moisture in a wet environment, combined with the risk of discharge caused by close proximity of conductors, representing the bird damage situation in high-risk seasons (spring rainy season, foggy days); modeling example: use the same bird model and position settings as the dry wading bird state; apply a higher conductivity to the surface of the bird geometry Such as 0.2 ~ 0.5 S / m), simulating the enhanced conductivity of feathers after absorbing moisture; at the same time, a layer of pollution coating is applied on the surface of the insulator, and the surface conductivity is defined as ~ S, used to simulate the surface leakage channel; set the equivalent humidity influence parameters for the air medium in the model (such as the air breakdown strength reduction ratio: The mesh encryption range includes not only the surrounding area of ​​the bird but also the possible paths between the bird and the insulator string. The simulation output is used to determine whether a conductive path is formed and whether the distance is close to the flashover threshold.

[0081] For the above working conditions, the air gap area between the wire and the armor is defined as the simulation solution domain. Based on the gap mesh created between the conductor and the armor structure, a hybrid mesh structure consisting of tetrahedral and hexahedral elements is further configured within this area. Local adaptive mesh refinement is automatically enabled at locations with drastic electric field changes, such as conductor edges and corners of insulating armor (when the rate of change of the local electric field gradient exceeds a preset curvature threshold), to ensure solution accuracy in areas with rapidly changing electric field strength.

[0082] S17: In the finite element simulation platform, after setting the boundary conditions, the electrostatic field solver is called for solution. The solution output includes: node potential function, electric field intensity vector and electric field intensity modulus in each grid cell based on the backcalculation of the potential gradient. ;

[0083] Furthermore, in each simulation condition, if the node meets the conditions where the electric field strength is greater than the electric field strength threshold (e.g., 4kV / cm) or the electric field gradient is greater than the electric field gradient critical value (e.g., 1.5×10³V / cm²), the node is marked as a pre-selected high-risk coordinate;

[0084] S18: Calculate the risk sensitivity weight for each pre-selected high-risk coordinate , when the risk sensitivity weight When , mark the pre-selected high-risk coordinate as the final high-risk coordinate. The calculation formula is:

[0085] ;

[0086] Where, is the potential difference between adjacent nodes; is the maximum operating voltage of the system; is the electric field gradient modulus in the current grid cell; is the critical breakdown field strength of air; is the safety gap threshold; is the Euclidean distance from the node to the nearest conductor; is the insulation armor protection factor, where , is the dielectric constant of the insulating armor material; 、 、 、 is the weight coefficient, which is obtained by training based on historical bird-related fault data;

[0087] S19: Store the electric field strength, electric field gradient, and final high-risk coordinates of all nodes to form an electric field distortion feature library.

[0088] S2: Based on the electric field distortion feature library, the spatial interference of bird bodies, nests, and excrement within the safe gap threshold is simulated, and the corresponding discharge probability and electric field distortion parameters are output. A generative adversarial network is used to synthesize bird images with category, simulated coordinates, lighting, weather, and posture information annotated, among which the lighting, weather, and posture information are used to control the synthesis of bird images.

[0089] In the embodiment of the present invention, the spatial interference of bird bodies, bird nests, and excrement within the safety gap threshold is simulated based on the electric field distortion feature library, and the specific implementation process of outputting the corresponding discharge probability and electric field distortion parameters is as follows:

[0090] S21: Based on the pre-built electric field distortion feature library, the gap area between the conductor and the conductive structure in the three-dimensional transmission tower gap model is obtained, and the minimum spacing between the conductor and the conductive structure in each grid cell within the gap area is calculated. Based on the set safe gap threshold, all grid cells with spacing less than the threshold are screened out, and a set of interference injection areas is constructed. In the interference injection area set, a set of simulation injection points is arranged according to multiple sets of discrete spatial coordinates for subsequent insertion of the foreign object model.

[0091] S22: For each injection point in the simulation injection point set, perform the following operations:

[0092] Insert bird body foreign body model: It is a long ellipsoid geometry, and its long axis direction can be arranged parallel or perpendicular to the wire. The conductivity parameter is selected according to the bird wading condition type. For example, in the dry bird wading condition, the conductivity is set to ; Under wet wading bird conditions, the conductivity is set to ;

[0093] Inserting foreign body model into bird nest: It adopts polyhedron shell structure, the shape of which can simulate the common bird nesting form. Its conductivity parameters are derived from the predefined material property library, such as hay material is set as , mixed branches / thin wire material is ;

[0094] Insertion of excrement foreign body model: It is a hemispherical droplet structure, and its conductivity adopts a humidity-related model, which is defined as: ,in is the ambient humidity, The reference conductivity is, for example, the volume conductivity of a saturated salt solution of bird excreta under standard environmental conditions (temperature 25°C, humidity 60% RH);

[0095] S23: After embedding the three types of foreign body models into the three-dimensional transmission tower gap model according to the injection point coordinates, the original boundary conditions, voltage levels, and material properties remain unchanged. The electrostatic field solver is called to re-solve the potential distribution after interference, output the electric field strength and electric field gradient, and calculate the corresponding discharge probability. , the calculation formula is: ;in, is the modulus of electric field intensity; is the critical breakdown field strength of air (the default value in this formula is 3kV / mm); is the slope adjustment parameter.

[0096] The discharge probability, electric field intensity, and electric field gradient output from each simulation injection point are recorded as a set of spatial disturbance feature data, which can be used as input for subsequent training sample generation or risk trend assessment.

[0097] S24: Use the Conditional Generative Adversarial Network (cGAN) structure to synthesize bird images to improve image structure control accuracy and physical consistency. The generation conditions include:

[0098] Type of bird target: such as crow, pigeon, eagle, etc.;

[0099] Simulation coordinates: The three-dimensional space coordinates (x, y, z) that correspond one-to-one to the injection points of the electric field disturbance simulation;

[0100] Light level: set values ​​such as morning light (L1), sunny noon (L2), evening backlight (L3);

[0101] Weather type: such as sunny, rainy, foggy, hot and humid, etc.;

[0102] Posture information: including wing opening and closing degree, head and neck angle, etc.

[0103] The bird's posture information is represented by a two-dimensional skeleton key point matrix, as follows:

[0104] Modeling the bird skeleton as a group The structure composed of two-dimensional image plane coordinate key points is recorded as: Commonly used bone key points include: beak tip, neck base, trunk center, left and right wing joints, wing ends, tail, claws, etc. The recommended value range is 8–12;

[0105] The skeleton keypoint matrix It is obtained by projecting and transforming a predefined three-dimensional bird model at different posture angles, or generated by real bird motion capture data; during the GAN training stage, the skeleton structure is input into the generator network in the form of a two-dimensional heat map, and participates in the image synthesis control process together with other conditional labels.

[0106] S3: Associate the discharge probability, electric field distortion parameters and bird images according to the simulated coordinates to form a training data set containing the conductivity of bird nest materials and the falling speed of excrement, which is used to train the risk decision model.

[0107] After the image is synthesized, the actual center coordinates of the bird target in the generated image are extracted (YOLO or skeleton tracking model). The spatial offset is compared with the corresponding input simulated coordinates based on the Euclidean distance calculation method. When the spatial offset is less than the set tolerance range (for example, 5 pixels), the image-label consistency is confirmed.

[0108] After the image-simulation consistency judgment is completed, the synthesized bird image is bound to the corresponding discharge probability, electric field intensity, and electric field gradient value through simulation coordinates to construct associated data entries that correspond one-to-one between the image and the physical disturbance characteristics, which are used to generate a structured training dataset.

[0109] The data structure of each sample in the training dataset is uniformly organized as a multi-dimensional input vector, and the output can be composed of the following fields:

[0110] Image encoding vectors of bird images, the encoding vectors are extracted by deep convolutional neural networks;

[0111] Electric field disturbance simulation parameters, including discharge probability, electric field intensity and electric field gradient;

[0112] Electrical conductivity of the nest material;

[0113] Fecal landing speed;

[0114] Simulate coordinate space location information.

[0115] S4: Obtain the tower monitoring video stream and input it into the target detection model that integrates the global attention mechanism and the SPPCSPC module for target recognition, and output the target category, real-time spatial coordinates and confidence.

[0116] In a further embodiment of the present invention, the target detection model is constructed based on a deep neural network structure. A global attention mechanism is integrated into the backbone network, and the global attention mechanism consists of a channel attention submodule and a spatial attention submodule;

[0117] The channel attention submodule performs global average pooling and global maximum pooling operations on the input feature map to generate two channel description vectors respectively. The two channel description vectors are encoded through a multi-layer perceptron (MLP) with shared weights, and the channel attention weights are output. The channel-level attention weights are fused with the original feature map by channel-by-channel multiplication to generate a channel-enhanced feature map.

[0118] The spatial attention submodule receives the occlusion area mask set based on the tower structure configuration. The mask is generated according to the preset template of the tower structure and marks the areas where the tower body, crossarms and insulators are located. The occlusion area mask participates in the spatial attention weight calculation and is used to perform weighted processing on the spatial response in the above-mentioned interference area to obtain an enhanced spatial perception feature map.

[0119] The SPPCSPC (Spatial Pyramid Pooling - Cross Stage Partial Connections) module is nested in the middle of the backbone feature extraction. SPPCSPC is a feature extraction module that integrates the spatial pyramid pooling structure (SPP) and the cross-stage partial connection structure (CSP). It includes multiple parallel convolution branches with different convolution kernel sizes (such as 1×1, 3×3, 5×5) and a set of channel fusion layers for connecting contextual features; the output feature map is concatenated and then compressed in dimension through the channel fusion convolution layer to extract multi-scale contextual semantic information.

[0120] The target detection model detects and temporarily caches the continuous image frames in the tower monitoring video stream frame by frame, records the center coordinates and confidence, and stores them in a fixed-size sliding time window; in several adjacent frames, the center point position of the same target is compared and the coordinate deviation is calculated. When the threshold stability requirements are met, a majority frame vote is performed to determine the final output target category, image center coordinates and confidence; the two-dimensional coordinates in the image frame are back-projected through the camera intrinsic parameter matrix and extrinsic parameter parameters to be mapped into three-dimensional real-time spatial coordinates.

[0121] Furthermore, the target detection model uses the Focal-EIoU loss function as the bounding box regression loss, and the formula is:

[0122] ;

[0123] Where IoU is the intersection-over-union ratio between the predicted box and the true box; is the focusing coefficient; To expand the IoU loss, it is used to comprehensively consider the overlap degree, center distance, and width and height difference between the predicted box and the real box; by giving greater weight to high IoU samples, this loss function balances the sample imbalance during training, improving the small target positioning accuracy and model convergence speed.

[0124] S5: Prioritize mapping the real-time spatial coordinates to the high-risk coordinate area of ​​the electric field distortion feature library, extract the corresponding electric field strength and gradient, and combine them with the target category, confidence, the conductivity of the bird's nest material, and the falling speed of the excrement to input the trained risk decision model. When the real-time spatial coordinates enter the safety gap threshold, the discharge risk score and breakdown trend indicator are calculated through the LSTM network.

[0125] like Figure 3As shown, in a further embodiment of the present invention, in the process of matching the real-time spatial coordinates output from the target detection model with the high-risk spatial coordinates in the electric field distortion feature library, a spatial index structure based on three-dimensional Euclidean distance is constructed to achieve preferential projection of target coordinates to high-risk areas and extract relevant electric field parameters. The method includes:

[0126] For high-risk coordinates in the electric field distortion feature library, a spherical neighbor search structure is constructed using a three-dimensional spatial indexing method, and the spherical search radius is defined using the Euclidean distance metric. , forming the following query mechanism:

[0127] In any real-time space coordinates For the query point, retrieve all High-risk coordinates If there are multiple matching points, select the nearest high-risk coordinate point ;

[0128] When there are high-risk coordinate points that meet the above conditions When the electric field intensity and gradient corresponding to the coordinate are extracted, the electric field intensity and gradient corresponding to the coordinate are used as the current real-time space coordinates. The associated electrical disturbance parameters;

[0129] If no matching coordinates that meet the distance condition are found among all high-risk coordinates, that is, all , then perform the global minimum distance search operation: in the entire simulation coordinate set in the electric field distortion feature library, calculate for each coordinate , select the coordinate point with the smallest distance , get The corresponding electric field strength and gradient are then used to label the sample as a "non-high-risk associated sample." This label is used as an auxiliary training input for the subsequent risk decision model, enabling the model to identify the characteristic boundary differences between high-risk locations and ordinary locations.

[0130] To improve computational efficiency, the spatial index structure in this embodiment can be implemented using general three-dimensional spatial neighbor structures such as KD-Tree, Ball-Tree, Octree, or Grid Hashing, and the search radius 𝑟 can be set to 0.1–0.3 meters according to the actual simulation point density.

[0131] In one embodiment of the present invention, the risk decision model is constructed based on a labeled training dataset derived from image recognition results, electric field simulation outputs, spatial matching operations, and physical parameter modeling. To enhance the model's ability to learn the interactions between spatial disturbances, bird behavior, and electric field distortion, the model input adopts a fused vector structure, divided into static input fields and sequential input fields.

[0132] Static input fields include: target category (such as coding labels such as crow, eagle, pigeon, etc.), confidence (confidence score output by the target detection model, range [0,1]), conductivity of bird nest material (unit S / m, from the material property library or measured label); time series input fields (organized by time frame number) include: real-time spatial coordinates, excrement landing speed (unit m / s, estimated by height and landing time), electric field strength (unit kV / cm, extracted from simulation), electric field gradient (unit kV / cm 2 ).

[0133] The risk decision model consists of two processing branches: a static feature processing branch and a temporal sequence processing branch. The static feature processing branch receives static input fields, converts them into fixed-length encoding vectors, and processes them through at least one fully connected layer and an activation function to generate a low-dimensional vector representation of the static risk factors of birds.

[0134] The time series processing branch receives a multi-frame input sequence consisting of time series input fields. The input structure is a sequence of time steps of length T. Each step contains: spatial coordinates (3D), landing point velocity (1D), electric field strength and gradient (2D), for a total of 6 dimensions. This input is fed into an LSTM network to extract its time-dependent structure. The LSTM hidden layer output serves as a compressed representation of the time series.

[0135] The static expression vector and the temporal expression vector are concatenated to obtain a fused representation vector. This vector is input to two independent output branches, and the fully connected output layer generates:

[0136] Discharge risk score (output range [0,1], indicating the current disturbance discharge probability score);

[0137] Breakdown trend indicator (the output range can be positive real numbers, indicating the current trend strength, and can be used as a continuous quantity to participate in warning level judgment).

[0138] The model training uses a supervised learning method, adopts the real discharge event annotations in the historical data as the target output, and defines the loss function for the two output values ​​for joint optimization.

[0139] S6: When the discharge risk score or breakdown trend index exceeds the dynamically adjusted warning threshold, a warning signal and fault coordinates are output.

[0140] In one embodiment of the present invention, in order to enhance the adaptive ability of the bird-related fault warning mechanism under different bird activity densities, spatial distributions and historical risk conditions, the system adopts a dynamically adjusted warning threshold strategy for the discharge risk scoring threshold. and breakdown trend indicator thresholds Set up a threshold update mechanism.

[0141] In the early stage of system deployment, the initial discharge risk score threshold is set based on expert experience or historical experimental data with the Breakdown Trend Indicator Threshold ;

[0142] During system operation, the target category distribution, real-time spatial coordinate distribution density, and historical high-risk sample triggering times within a unit time window are counted, and dynamic threshold adjustment is performed every sliding time window (such as 30 seconds or 1 minute). In each window period, the system counts the following parameters and calculates the dynamic adjustment factor:

[0143] Category risk factors : Based on the bird target categories detected in the window, the high-risk species with the highest frequency of occurrence (such as large birds of prey and gregarious birds) are counted, and a risk weight is set for each category (such as 1.2 for eagles and 1.0 for pigeons). The weight coefficient is generated comprehensively: ,in is the category risk coefficient, is the number of occurrences of this type, is the total number of detection targets;

[0144] Spatial interference density factor : Counts the percentage of times that the real-time spatial coordinates fall within the "safety gap threshold" range, defined as: ,in Indicates the target number of times that the gap falls within the range. is the total number of all detection targets;

[0145] Historical trigger rate factor : The system maintains the discharge risk score and actual discharge event trigger frequency within a period (such as the past 10 minutes), which are defined using a normalized counting method: ,in, The number of high-risk events triggered. is the total number of samples in this time period;

[0146] Combine the three types of adjustment factors in a multiplicative manner and use the adjustment rate coefficient (For example ) controls the update intensity and uses the exponential decay formula to calculate the updated threshold:

[0147] ;

[0148] ;

[0149] Where, is the dynamic adjustment coefficient, with a value range of [0,1], used to control the adjustment rate; 、 They are time, The discharge risk score threshold at the moment; 、 They are time, Breakdown trend indicator threshold at the moment.

[0150] The threshold update will be lower bounded after each iteration, e.g. , to avoid an increase in false positives caused by a too low threshold.

[0151] The early warning signal includes sending an alarm prompt to the monitoring center and control instructions for the on-site linkage device: when the assessed risk exceeds the threshold, the system generates an information alarm containing the tower number, risk type and level, which is displayed in a pop-up window on the background monitoring interface and recorded in a log; at the same time, the on-site bird-repellent equipment is triggered to emit sound, light or ultrasonic signals to drive away birds, and the early warning event is notified to the operation and maintenance personnel through the communication module so that further manual intervention measures can be taken in time.

[0152] Furthermore, the method also includes a fault-tolerant mechanism, specifically: when there is a missing video stream of the tower monitoring, millimeter-wave radar data is obtained, and the radar spatial coordinates are extracted to replace the real-time spatial coordinates; when the electric field distortion parameters corresponding to the real-time spatial coordinates are missing in the electric field distortion feature library, the training data set is called to interpolate and compensate for the discharge probability.

[0153] In summary, the present invention takes into account simulation accuracy, real-time identification and scientific risk assessment, and has good engineering adaptability and promotion prospects.

[0154] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A bird-related fault identification and early warning method based on a transmission tower clearance model, characterized in that: The method comprises: Construct a three-dimensional transmission tower gap model with partial insulation armor, calculate the electric field distribution of the three-dimensional transmission tower gap model under no-load, dry bird-wading conditions, and wet bird-wading conditions through finite element simulation, and output an electric field distortion feature library, including electric field intensity, gradient, and high-risk coordinates; Based on the electric field distortion feature library, the spatial interference of bird bodies, nests, and excrement within a safe gap threshold is simulated, and the corresponding discharge probability and electric field distortion parameters are output. A generative adversarial network is then used to synthesize bird images annotated with category, simulated coordinates, lighting, weather, and posture information, where the lighting, weather, and posture information are used to control the synthesis of the bird images. Associating the discharge probability and electric field distortion parameters with the bird image according to the simulated coordinates to form a training data set containing the conductivity of the bird's nest material and the falling speed of the excrement, which is used to train the risk decision model; Obtain tower monitoring video streams and input them into the target detection model that integrates the global attention mechanism and the SPPCSPC module for target recognition, outputting target categories, real-time spatial coordinates, and confidence levels; Prioritizing the mapping of the real-time spatial coordinates to the high-risk coordinate area of ​​the electric field distortion feature library, extracting the corresponding electric field strength and gradient, and combining them with the target category, confidence level, the conductivity of the bird's nest material, and the falling speed of the excrement, inputting them into the trained risk decision model. When the real-time spatial coordinates enter the safe gap threshold, the discharge risk score and breakdown trend indicator are calculated through the LSTM network; When the discharge risk score or breakdown trend index exceeds a dynamically adjusted warning threshold, a warning signal and fault coordinates are output.

2. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 1 is characterized in that: The method also includes a fault-tolerant mechanism, specifically: when there is a missing video stream of the tower monitoring, millimeter-wave radar data is obtained, and the radar spatial coordinates are extracted to replace the real-time spatial coordinates; when the electric field distortion parameters corresponding to the real-time spatial coordinates are missing in the electric field distortion feature library, the training data set is called to interpolate and compensate for the discharge probability.

3. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 1 is characterized in that: The construction of a three-dimensional transmission tower gap model with local insulation armor specifically includes: Input the geometric parameters of the tower body, crossarms, insulators, conductor installation positions, and insulation armor structure to generate a 3D structural model including a complete space frame. Input the electrical parameter set and the physical parameters of the materials of each structural component, establish a material property combination, and bind the material property combination to the unit ID of the space frame one by one; Constructing an insulating armor subunit nested in the tower body and the crossarm through a node coupling algorithm, and incorporating the material property combination; Call the predefined transmission tower structure template library and complete automatic instantiation and combination based on structural parameters; Grid cells are divided in the gap area between the conductor and the conductive structure, and the spatial frame, electrical parameter set and material property combination are assigned to each grid cell to construct a three-dimensional transmission tower gap model with local insulating armor for simulation.

4. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 3 is characterized in that: The electric field distribution of the three-dimensional transmission tower gap model under no-load, dry bird-wading conditions, and wet bird-wading conditions is calculated by finite element simulation, and the method includes: Based on the divided grid cells, the gap area between the wire and the conductive structure is reconstructed into a hybrid network composed of tetrahedral cells and hexahedral cells. When the local electric field gradient change rate exceeds the preset curvature threshold, adaptive grid refinement is triggered. The electrostatic field solver is used for no-load, dry wading bird, and wet wading bird conditions to calculate the node potential distribution and output the electric field strength and gradient of each node; If the node meets the condition that the electric field strength is greater than the electric field strength threshold or the electric field gradient is greater than the electric field gradient critical value, the node is marked as a pre-selected high-risk coordinate; For each pre-selected high-risk coordinate, calculate the risk sensitivity weight , the calculation formula is: ； Where, is the potential difference between adjacent nodes; is the maximum operating voltage of the system; is the electric field gradient modulus; is the critical breakdown field strength of air; is the safety gap threshold; is the Euclidean distance from the node to the nearest conductor; is the insulation armor protection factor, where , is the dielectric constant of the insulating armor material; 、 、 、 is the weight coefficient; When the risk sensitivity weight When , mark the pre-selected high-risk coordinate as the final high-risk coordinate; The electric field intensity, electric field gradient and final high-risk coordinates of all nodes are stored to form an electric field distortion feature library.

5. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 1 is characterized in that: The method of simulating the spatial interference of a bird body, a bird nest, and excrement within a safe gap threshold based on the electric field distortion feature library and outputting corresponding discharge probability and electric field distortion parameters includes: Extracting grid cells where the gap distance between the conductor and the conductive structure is less than a set safety gap threshold from the electric field distortion feature library, constructing an interference injection area set, and arranging a simulation injection point set in the interference injection area set according to multiple sets of discrete spatial coordinates; For each injection point in the simulation injection point set, three types of foreign body models, bird body, bird nest, and excrement, are inserted respectively. The bird body foreign body model is a long ellipsoid geometry, and the conductivity parameters are selected according to the dry wading bird working condition and the wet wading bird working condition respectively. The bird nest foreign body model adopts a polyhedron shell structure, and the conductivity parameters are derived from the predefined material property library. The excrement foreign body model is a hemispherical droplet structure, and the conductivity is ,in is the ambient humidity, is the reference conductivity; After embedding the three types of foreign body models into the three-dimensional transmission tower gap model according to the injection point coordinates, the electrostatic field solver is called to re-solve the potential distribution after interference, output the electric field strength and gradient, and calculate the corresponding discharge probability in combination with the air breakdown strength parameter. , the formula is: ;in, is the modulus of electric field intensity; is the critical breakdown field strength of air; is the slope adjustment parameter.

6. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 5 is characterized in that: Methods for synthesizing bird images and forming training datasets include: A conditional generative adversarial network (CGN) is used to synthesize bird images. The generation conditions include the bird's category, simulated coordinates, light level, weather type, and posture information, including wing opening and closing and head and neck angle. After image synthesis, the actual center coordinates of the bird target in the generated image are extracted and compared with the corresponding input simulated coordinates based on the Euclidean distance calculation method. When the spatial offset is less than the set tolerance range, the image-label consistency is confirmed. After the image-simulation consistency judgment is completed, the synthesized bird image is bound to the corresponding discharge probability, electric field intensity, and electric field gradient value through simulation coordinates to construct associated data entries that correspond one-to-one between the image and the physical disturbance characteristics, which are used to generate a structured training dataset.

7. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 1 is characterized in that: The global attention mechanism includes a channel attention submodule and a spatial attention submodule; The channel attention submodule generates channel attention weights by performing global average pooling and maximum pooling operations on the input feature map, and performs channel-by-channel product fusion on the channel attention weights and the original feature map; the spatial attention submodule receives an occlusion area mask set based on the tower structure configuration, and the occlusion area mask is used to adjust the spatial response of the tower body, crossarm and insulator interference area in the spatial attention calculation; The SPPCSPC module is nested in the feature extraction path between the middle of the backbone network and the output feature layer, and includes multiple parallel convolution branches with different convolution kernel sizes and a set of channel fusion layers for connecting context features; The target detection model detects consecutive image frames in the tower monitoring video stream frame by frame, stores candidate target outputs in a fixed time window, compares the spatial positions of the same target in consecutive frames and calculates the deviation of the center point coordinates in the frame. When the set threshold conditions are met, a majority frame voting judgment is performed to determine the final output target category, center coordinates in the image frame and confidence level; the center coordinates in the image frame are mapped into three-dimensional real-time spatial coordinates after camera parameter transformation.

8. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 1 is characterized in that: The real-time spatial coordinates are preferentially mapped to the high-risk coordinate area of ​​the electric field distortion feature library, and the corresponding electric field strength and gradient are extracted. The method includes: A spatial index structure is constructed for high-risk coordinates in the electric field distortion feature library. The spatial index structure is constructed based on the three-dimensional Euclidean distance calculation with a radius of The spherical search domain; The real-time spatial coordinates are input into the spatial index structure, and a spatial matching operation is performed. When the distance between any high-risk coordinate and the real-time spatial coordinate is less than or equal to When , the electric field intensity and gradient corresponding to the high-risk coordinate are extracted; If the real-time space coordinate does not fall within the tolerance search domain of any high-risk coordinate, a global minimum distance search is performed between the real-time space coordinate and all coordinates in the electric field distortion feature library to obtain the electric field strength and gradient corresponding to the nearest neighbor coordinate, and the data sample is marked as a "non-high-risk associated sample" for model-assisted input.

9. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 1 is characterized in that: The risk decision model constructs a fusion input vector based on the training data set, and divides the fusion input vector into a static input field and a time series input field. The static input field includes the target category, confidence level, and the conductivity of the bird's nest material, and the time series input field includes the real-time spatial coordinates, the speed of the excrement landing point, the electric field strength and gradient; The risk decision model includes a static feature processing branch and a time series processing branch, wherein the static feature processing branch is a subnetwork including at least one fully connected layer, which is used to extract static feature expressions from static input fields; The time series processing branch is an LSTM network, which is used to process time series input fields and output time series feature expressions; The static feature expression and the time series feature expression are spliced ​​and fused, and two output values ​​are generated through a fully connected layer, which are a discharge risk score and a breakdown trend indicator.

10. The bird-related fault identification and early warning method based on the transmission tower clearance model according to claim 1 is characterized in that: The dynamic adjustment method of the warning threshold is: Setting the initial discharge risk score threshold with the Breakdown Trend Indicator Threshold ; During the operation of the system, the target category distribution, real-time spatial coordinate distribution density and historical high-risk sample triggering times within the unit time window are counted, and the category risk factor is calculated based on the high-risk bird category with the highest current occurrence frequency. , calculate the spatial interference density factor based on the frequency change of spatial coordinates falling within the safety gap threshold , calculate the historical trigger rate factor based on the actual discharge events or high confidence predictions within the past preset time period ; The dynamically adjusted warning threshold is calculated using the weighted update formula: ； ； Where, is the dynamic adjustment coefficient, with a value range of [0,1], used to control the adjustment rate; 、 They are time, The discharge risk score threshold at the moment; 、 They are time, Breakdown trend indicator threshold at the moment.

Images