Intelligent detection method for outdoor power line fault detection

Abstract

The invention discloses an intelligent detection method for fault detection of an outdoor power line, and the method comprises the following steps: 1, carrying out the collection and preprocessing of multi-modal data, and carrying out the collection and preprocessing of the multi-modal data through an unmanned plane cluster, a distributed optical fiber sensor, a laser radar and meteorological monitoring equipment; visible light image data, infrared image data, laser point cloud data, vibration waveforms, temperature distribution and environmental parameters of the power line are synchronously obtained, and multi-source image data are processed, namely the visible light image data, the infrared image data and the laser point cloud data are processed; and 2, intelligent fault diagnosis: inputting the data acquired in the step 1 into a multi-task neural network model, and outputting a fault positioning and type identification result. According to the novel detection method based on multi-modal data fusion, an intelligent algorithm and closed-loop optimization, the fault identification precision, the dynamic decision-making capability and the comprehensive protection efficiency are improved, and the intelligent operation and maintenance requirements of a modern power grid are met.

Description

Technical Field

[0001] The present invention relates to the field of power detection, and in particular to an intelligent detection method for outdoor power line fault detection. Background Art

[0002] The operating environment of outdoor power lines is complex and the types of faults are diverse. Outdoor power lines are exposed to the natural environment for a long time and face multiple threats such as meteorological changes, such as strong winds and heavy rains, complex terrain, such as mountainous areas and forested areas, and biological activities, such as bird nesting and rodent gnawing. Outdoor power lines are subject to short circuits, disconnections, overheating, insulation deterioration, abnormal mechanical stress, etc. Traditional methods cannot quickly and accurately identify the root causes of various types of faults.

[0003] Traditional detection methods have significant limitations:

[0004] Manual inspection: This method relies on manual visual inspection, which is inefficient, has limited coverage, and poses safety risks for high-altitude operations. It is difficult to adapt to the real-time monitoring needs of large-scale line networks.

[0005] Single sensor monitoring: Using only a single type of sensor, such as infrared or vibration sensors, is unable to integrate multi-dimensional data, such as visible light images, laser point clouds, and temperature fields. This results in incomplete fault feature extraction and is prone to missed or misjudgment.

[0006] Static decision-making model: Static analysis based on fixed thresholds or historical experience, lacks the ability to respond in real time to dynamic environments, such as changes in the biological activity threat index and fluctuations in meteorological parameters, and the disposal plan lacks flexibility and scientificity.

[0007] Therefore, an intelligent detection method for outdoor power line fault detection is proposed. Summary of the Invention

[0008] In view of the defects in the prior art, the present invention provides an intelligent detection method for outdoor power line fault detection, comprising the following steps:

[0009] Step 1: Multimodal data acquisition and preprocessing: Using drone clusters, distributed fiber optic sensors, lidar, and meteorological monitoring equipment, we simultaneously acquire visible light image data, infrared image data, laser point cloud data, vibration waveforms, temperature distribution, and environmental parameters of power lines. We then process the multi-source image data, i.e., visible light image data, infrared image data, and laser point cloud data.

[0010] Step 2: Intelligent fault diagnosis: The multimodal data preprocessed in Step 1 is fed into a multi-task neural network model, which outputs fault location and type identification results. The multi-task neural network model is a cascaded structure of a convolutional neural network based on ResNet-50 and an LSTM network. The training data is a historical dataset containing short circuit, disconnection, and overheating faults, and the loss function is cross-entropy loss.

[0011] Step 3: Dynamic decision generation: Based on the analysis of fault type and environmental parameters, a decision plan is generated that includes handling priority, risk level, and resource scheduling;

[0012] Step 4: Autonomous disposal execution, calling intelligent operation equipment according to the fault type;

[0013] Step 5: Closed-loop optimization feedback: collect and process the data from steps 1 to 4, update the fault handling knowledge graph, and optimize the decision-making strategy through reinforcement learning.

[0014] Furthermore, the specific process of step 2 includes:

[0015] Construct spatiotemporal feature tensor:

[0016]

[0017] Where: Concat is a feature concatenation operation that concatenates features of different modalities in the channel dimension;

[0018] F img is the two-dimensional spatial feature matrix extracted from the visible light / infrared image, with dimensions of H×W, representing the image height and width;

[0019] F vib is the vibration spectrum feature vector converted from distributed optical fiber vibration data, and its dimension is related to the frequency band division;

[0020] F temp Generate a temperature field distribution map for the temperature sensor network with a spatial resolution that matches the sensor density;

[0021] H and W are the height and width of the feature map;

[0022] C is the total number of feature channels after fusion;

[0023] Extract cross-modal correlation features through multi-scale convolution kernels;

[0024] Output failure probability matrix:

[0025]

[0026] P(y|x) is the fault probability matrix distribution, which represents the probability of fault type y under the input data x. W is the classification weight matrix, which is used to map high-dimensional features to the fault category space.

[0027] b is the bias vector, which is used to adjust the classification decision boundary;

[0028] is the spatiotemporal feature tensor Spatial pooling operation to compress the spatial dimension (H×W);

[0029] W is the classification matrix with a dimension of K×D, where K is the total number of fault types and D is the feature dimension after pooling;

[0030] b is a bias vector with a dimension of K×1, which is used to adjust the classification decision boundary.

[0031] Furthermore, the environmental status analysis in step 3 includes a biological activity risk assessment step, the specific process of which is:

[0032] Detect biological aggregation areas around the line through visible light-thermal imaging fusion;

[0033] Use voiceprint recognition algorithms to analyze biological species characteristics;

[0034] Calculate the biological activity threat index. The specific process is as follows:

[0035]

[0036] W i is the preset hazard weight coefficient for the i-th species. It ranges from [0,1] and is defined by the species risk level table. It is used to quantify the difference in threats posed by different species to electrical equipment.

[0037] A i is the area covered by biological activities of type i, which is used to reflect the scale of biological aggregation;

[0038] A0 is the baseline dangerous area, which is set based on historical accident statistics, such as the Bird's Nest A0 = 0.5m 2 , used to standardize the area impact factor;

[0039] t i The duration of biological activity, used to characterize the cumulative effect of threats over time;

[0040] k is the time influencing factor, which controls the exponential growth rate and is calibrated by the species behavior pattern to adjust the sensitivity of time to threats;

[0041] It is an exponential function used to amplify the contribution of long-term biological activities to the threat and reflect the cumulative effect of the threat;

[0042] n is the total number of species, that is, the number of different biological categories detected, such as birds, rodents, insects, etc.

[0043] i is the loop variable, representing the index of the currently calculated biological category, i = 1, 2, ..., n.

[0044] Furthermore, after obtaining the biological activity threat index Tb, when the biological activity threat index Tb exceeds a preset threshold, a biological protection disposal plan is generated, and the biological protection disposal plan includes:

[0045] Build a multimodal processing instruction set, including:

[0046] Sonic repellent mode: Dynamically adjust the frequency of sonic waves according to the characteristics of the species, specifically:

[0047]

[0048] f0 is the reference repellent frequency, which is set according to the biological auditory sensitivity frequency band, such as f0 = 3kHz for birds, and is used to provide basic repellent capability;

[0049] η is the adjustment coefficient, ranging from [0.2 to 0.8], which is used to control the amplitude of the frequency change with the threat;

[0050] T b is the biological activity threat index;

[0051] T th is the threat threshold, used to avoid over-response to low-threat scenarios;

[0052] T max It is the maximum threat value, the upper limit of system design, used to limit the frequency adjustment range;

[0053] Physical protection mode: The insulation protection net is deployed through the coordinated deployment of drone swarms. The mesh density is as follows:

[0054]

[0055] ρ min It is the minimum permissible mesh density, set by insulation requirements, to ensure basic protection performance;

[0056] ρ max It is the maximum permissible mesh density, limited by the material strength, used to prevent structural failure;

[0057] A crit It is the critical protection area, set according to the line voltage level, and is used to dynamically balance the protection strength and ecological permeability;

[0058] A i The activity coverage area of ​​type i organisms;

[0059] Ecological transfer model: deploying intelligent traps to release species-specific pheromones with concentration gradients that meet the following requirements:

[0060]

[0061] is the pheromone concentration gradient, which is used to guide the directional migration of organisms. κ is the diffusion coefficient, which is calibrated by the ambient temperature and humidity to match the biological perception characteristics. is the rate of change of the biological threat index over time, reflecting the dynamic growth or decay of the threat;

[0062] Establish a real-time evaluation model for disposal effects, including:

[0063] The effectiveness indicators of acoustic repellent are as follows:

[0064] E s To quantify the effectiveness of acoustic repellent and the duration of the repellent effect;

[0065] ΔS post is the reduction in biological activity area after treatment;

[0066] ΔS pre is the original value of the biological activity area before disposal;

[0067] β is the decay coefficient, which controls the rate at which effectiveness decreases over time;

[0068] t is time;

[0069] The protection net coverage index is as follows:

[0070] M detected is the set of detected biological activity areas;

[0071] M covered It is the set of areas actually covered by the protection net;

[0072] Dynamically switch the treatment mode combination according to the real-time evaluation model when: When Cc is less than δ, δ is the preset value, and the multi-mode collaborative disposal control is started. The control logic is:

[0073]

[0074] e(t) is the threat deviation, that is, e(t) = Tb(t) - T targete , which is the difference between the current threat and the target threat;

[0075] K pis the proportional gain, ranging from [0.5, 2.0], used to quickly respond to threat changes;

[0076] K d is the differential gain, ranging from [0.01, 0.1], used to suppress overshoot and oscillation;

[0077] K i is the integral gain, eliminating steady-state error;

[0078] K p e(t) is a proportional term that directly responds to the current threat deviation and quickly adjusts the treatment intensity;

[0079] is an integral term, which is used to accumulate historical deviations and eliminate long-term steady-state errors;

[0080] It is a differential term used to predict the threat change trend and suppress overshoot and oscillation.

[0081] Furthermore, the multi-source image data processing in step 1 includes:

[0082] Process the visible light image data, that is, perform equipment deformation detection through the visible light channel and calculate the displacement of the hardware:

[0083] Δd=||SIFT(I t )-SIFT(I t-1 )||2;

[0084] Δd is the displacement of hardware, which represents the deformation and displacement of power line equipment, such as insulators or connecting hardware, at adjacent time points;

[0085] SIFT(·) is a scale-invariant feature transform algorithm used to extract key points and their feature descriptors from images;

[0086] I t is the visible light image at the current time (t), used to capture the real-time status of the device;

[0087] I t-1 is the visible light image at the previous moment (t-1), which serves as the reference for deformation detection;

[0088] Process infrared image data, identify abnormal temperature rise areas through infrared channels, and locate overheating faults at connection points;

[0089] The laser point cloud data is processed, and line sag changes are analyzed through the laser point cloud to detect mechanical stress anomalies.

[0090] The multi-source image data processing in step 1 also involves fusing multi-period image data to construct a 3D degradation model and predict the remaining life of components. The specific process is as follows:

[0091] L r =L0·exp(-λ·∫σ(t)dt);

[0092] L0 is the initial design life of the component, which is used to provide a life prediction benchmark;

[0093] σ(t) is the time-varying stress function, which is used to quantify the impact of the external environment on life;

[0094] λ is the material degradation coefficient, which is calibrated through accelerated aging experiments and is used to reflect the material's fatigue resistance;

[0095] ζσ(t)dt is the time integral of the time-varying stress, which represents the cumulative external stress to which the component is subjected during its entire life cycle;

[0096] exp is an exponential function used to map the cumulative effect of external stress into a nonlinear relationship of life decay.

[0097] Furthermore, the specific process of step 4 includes:

[0098] Automatically select operating equipment based on fault type: when there is a mechanical fault, a climbing robot is called;

[0099] Insulation inspection drones are activated when there is an electrical fault;

[0100] Activate the sonic repellent device when there is a biologically related fault;

[0101] At the same time, a safe operation path is generated, and the objective function of safe operation path planning is:

[0102]

[0103] R risk The dynamic risk coefficient of the path segment is a 0-1 normalized value that combines the following risk factors:

[0104] Electrical risks: line voltage level, discharge probability, etc.

[0105] Biological risk: Threat index of surrounding biological activities.

[0106] Terrain risk: terrain complexity, such as slope and obstacle density.

[0107] Calculation method: Real-time evaluation through multimodal data to quantify the degree of path danger;

[0108] d iis the Euclidean length of the path segment, which represents the distance the robot arm or drone moves on the path segment. It is used to control the energy consumption of movement. The longer the path, the higher the energy consumption cost.

[0109] T total The total time estimate for completing a fault repair operation, including movement time and operation time, such as the time it takes for a climbing robot to replace a component, is used together with the risk cost to form a multi-objective optimization function to avoid inefficiency caused by excessive pursuit of low risk.

[0110] α is the time cost conversion coefficient, which is used to achieve multi-objective optimization.

[0111] Furthermore, the process of planning the safe operation path further includes:

[0112] Rehearse the disposal process through the digital twin system to detect conflicting solutions;

[0113] Dynamically adjust the robot arm's operating trajectory, specifically:

[0114]

[0115] q0 is the initial angle vector of the robot arm joint, which provides the starting reference position of the motion trajectory and serves as a reference point for trajectory adjustment to ensure that the robot arm starts operation from a known safe state;

[0116] K p is the proportional gain coefficient, which quickly responds to the current trajectory tracking error;

[0117] e(τ) is the trajectory tracking error, that is, the difference between the actual position and the target position, which is used as the basis for feedback control;

[0118] K i is the integral gain coefficient, which is used to eliminate the cumulative effect of historical errors;

[0119] ∫e(τ)dτ is the time integral of the error, which represents the accumulated error from the initial time to the current time t;

[0120] dτ is the differential symbol of time, indicating that the integral variable is time τ.

[0121] Furthermore, the specific contents of step five include:

[0122] Build a fault handling knowledge graph to associate fault characteristics, handling solutions, and effect indicators;

[0123] The decision-making strategy is optimized through reinforcement learning. The decision-making strategy function optimized by reinforcement learning is:

[0124]

[0125] T repairThe actual repair time is the time taken from the occurrence of a fault to complete repair, reflecting the handling efficiency. It is calculated based on real-time data such as the movement speed and operation complexity of intelligent operation equipment such as drones and robots;

[0126] T max The maximum allowable repair time is set by the power supply reliability standard, such as the urban power supply requirement T max = 120 minutes, used to standardize time efficiency indicators and ensure that the reward value matches the actual business needs;

[0127] C saved To recover the economic losses, we can calculate the cost by reducing the power outage time and the degree of equipment damage. Example: If the power outage lasts for 1 hour and causes a loss of 500,000 yuan, and the repair time is shortened to 0.5 hours, then C saved =250,000 yuan;

[0128] C total The maximum potential loss amount is set based on the average loss value of similar failures in the same period in history and is used to standardize the economic impact index. For example, if the historical average loss is 1 million yuan, then C total =1 million yuan;

[0129] ω1 and ω2 are time and economic weight coefficients, satisfying ω1+ω2=1, and are used to define the optimization target priority.

[0130] The beneficial effects of the present invention are as follows: through multimodal data fusion and multi-task neural network model, multi-dimensional perception and high-precision positioning of power line faults are realized, and complex fault types such as short circuit, disconnection, overheating, etc. are effectively identified. Based on real-time environmental status analysis and threat assessment, decision-making plans including priority, risk level, and resource scheduling are automatically generated to improve the scientificity and timeliness of emergency disposal. Disposal tasks are automatically executed by intelligent operation equipment, and combined with closed-loop optimization feedback mechanism, knowledge graphs and decision-making strategies are continuously updated to improve the system's adaptability and long-term operation stability. Voiceprint recognition, thermal imaging and dynamic threat index calculation are integrated to accurately Accurately assess threats from biological activities and adopt multimodal collaborative protection, acoustic repellent, protective nets, and ecological transfer to reduce the risk of line failures caused by biological factors. Through digital twin rehearsals and safe path planning, the operating trajectories of robotic arms and drones are dynamically optimized to reduce the risk of manual intervention and ensure efficient and safe fault repair in complex environments. A three-dimensional degradation model is constructed based on multi-period image data to predict the remaining life of components, provide data support for preventive maintenance of power lines, and extend the service life of equipment. Combined with reinforcement learning and knowledge graph technology, the system can continuously accumulate historical cases and optimize strategies to adapt to diverse fault scenarios and new risk challenges. BRIEF DESCRIPTION OF THE DRAWINGS

[0131] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0132] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0133] The following embodiments of the technical solution of the present invention will be described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and are therefore only examples and are not intended to limit the scope of protection of the present invention.

[0134] It should be noted that, unless otherwise specified, the technical or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0135] like Figure 1 As shown, an intelligent detection method for outdoor power line fault detection includes the following steps:

[0136] Step 1: Multimodal data acquisition and preprocessing: Using drone clusters, distributed fiber optic sensors, lidar, and meteorological monitoring equipment, we simultaneously acquire visible light image data, infrared image data, laser point cloud data, vibration waveforms, temperature distribution, and environmental parameters of power lines. We then process the multi-source image data, i.e., visible light image data, infrared image data, and laser point cloud data.

[0137] Step 2: Intelligent fault diagnosis: The multimodal data preprocessed in Step 1 is fed into a multi-task neural network model, which outputs fault location and type identification results. The multi-task neural network model is a cascaded structure of a convolutional neural network based on ResNet-50 and an LSTM network. The training data is a historical dataset containing short circuit, disconnection, and overheating faults, and the loss function is cross-entropy loss.

[0138] Step 3: Dynamic decision generation: Based on the analysis of fault type and environmental parameters, a decision plan is generated that includes handling priority, risk level, and resource scheduling;

[0139] Step 4: Autonomous disposal execution, calling intelligent operation equipment according to the fault type;

[0140] Step 5: Closed-loop optimization feedback: collect and process the data from steps 1 to 4, update the fault handling knowledge graph, and optimize the decision-making strategy through reinforcement learning.

[0141] The specific process of step 2 includes:

[0142] Construct spatiotemporal feature tensor:

[0143]

[0144] Where: Concat is a feature concatenation operation that concatenates features of different modalities in the channel dimension;

[0145] F img is the two-dimensional spatial feature matrix extracted from the visible light / infrared image, with dimensions of H×W, representing the image height and width;

[0146] F vib is the vibration spectrum feature vector converted from distributed optical fiber vibration data, and its dimension is related to the frequency band division;

[0147] F temp A temperature field distribution map generated for a temperature sensor network with a spatial resolution matching the sensor density;

[0148] H and W are the height and width of the feature map;

[0149] C is the total number of feature channels after fusion, which is determined by the superposition of feature dimensions of each modality;

[0150] Extract cross-modal correlation features through multi-scale convolution kernels;

[0151] Output failure probability matrix:

[0152]

[0153] P(y|x) is the fault probability matrix distribution, which represents the probability of fault type y under the input data x. W is the classification weight matrix, which is used to map high-dimensional features to the fault category space.

[0154] b is the bias vector, which is used to adjust the classification decision boundary;

[0155] is the spatiotemporal feature tensor Spatial pooling operation to compress the spatial dimension (H×W);

[0156] W is the classification matrix with a dimension of K×D, where K is the total number of fault types and D is the feature dimension after pooling;

[0157] b is the bias vector with dimension K×1, which is used to adjust the classification decision boundary;

[0158] This process integrates heterogeneous data such as visible light images, vibration spectra, and temperature fields, breaking the limitations of a single sensor and enhancing the comprehensive extraction of fault features. Softmax classification maps complex features into the probability distribution of specific fault types, improving the reliability and interpretability of diagnostic results. Joint modeling of spatiotemporal features captures the spatiotemporal correlations of faults, such as the coupling effect between mechanical vibration and temperature anomalies, avoiding the one-sidedness of isolated analysis.

[0159] For example, if a transmission line vibrates abnormally due to loose connection hardware, the vibration spectrum characteristic F vib changes, and the local temperature rises, the temperature field F temp Abnormal. Traditional methods may use only a single data source, such as temperature, to misjudge an overheating fault. However, the method in this case:

[0160] Feature fusion: The visible light image F imb That is to capture the hardware displacement and vibration spectrum F vib That is to detect mechanical vibration and temperature field F temp That is, the temperature rise is monitored and spliced ​​into a spatiotemporal feature tensor, the correlation between mechanical looseness and temperature rise is comprehensively identified, and the root cause of the problem, i.e., loose hardware, is directly located through fault probability matrix calculation to avoid misjudgment.

[0161] The environmental status analysis in step 3 includes a biological activity risk assessment step, and the specific process is as follows:

[0162] Detect biological aggregation areas around the line through visible light-thermal imaging fusion;

[0163] Use voiceprint recognition algorithms to analyze biological species characteristics;

[0164] Calculate the biological activity threat index. The specific process is as follows:

[0165]

[0166] w i is the preset hazard weight coefficient of the i-th species, ranging from [0,1], defined by the species risk level table, used to quantify the difference in threats posed by different species to power equipment;

[0167] A i is the area covered by biological activities of type i, which is used to reflect the scale of biological aggregation;

[0168] A0 is the baseline dangerous area, which is set based on historical accident statistics, such as the Bird's Nest A0 = 0.5m 2 , used to standardize the area impact factor;

[0169] t i The duration of biological activity, used to characterize the cumulative effect of threats over time;

[0170] k is the time impact factor, which is used to adjust the sensitivity of time to threats;

[0171] It is an exponential function used to amplify the contribution of long-term biological activities to the threat and reflect the cumulative global effect of the threat;

[0172] n is the total number of species, that is, the number of different biological categories detected, such as birds, rodents, insects, etc.;

[0173] i is the loop variable, which represents the index of the biological category currently being calculated, i = 1, 2, ..., n;

[0174] Through the biological activity risk assessment and multimodal protection coordination mechanism, accurate threat quantification is achieved, combining visible light-thermal imaging fusion, voiceprint recognition and dynamic threat index T b , scientifically quantify the risks of biological activities, avoid errors in subjective experience judgment, and automatically match the treatment mode according to the characteristics of biological species, such as birds and rodents, namely acoustic repellent, protective net or ecological transfer, to improve the targeted protection and eco-friendliness, evaluate the treatment effect in real time, and dynamically switch between multiple mode combinations through PID control logic to ensure the continuous optimization of protection measures;

[0175] For example, if squirrels are active frequently near a high-voltage line, their gnawing behavior may cause a short circuit.

[0176] The system detects dense biological heat sources through thermal imaging, covering area A i =2m 2 , combined with the preset hazard weight w i =0.8, activity duration t i = 5 days, calculate T b ≈6.7, exceeding the threshold, triggering acoustic repellent, and at the same time the drone deploys an insulating protective net. If the effectiveness of the acoustic wave is detected to decrease, that is, the repellent effect is weakened and the protection net coverage index does not meet the standard, the system automatically starts the ecological transfer mode, that is, releasing squirrel pheromones, and ultimately driving away the creatures and eliminating the risk.

[0177] This allows for precise location of the threat source, avoiding traditional single-mode expulsion methods, such as the interference of fixed-frequency sound waves on non-target organisms such as birds, while significantly reducing the probability of line failure through multi-mode collaboration.

[0178] After obtaining the biological activity threat index Tb, when the biological activity threat index Tb exceeds the preset threshold, a biological protection disposal plan is generated. The biological protection disposal plan includes:

[0179] Build a multimodal processing instruction set, including:

[0180] Sonic repellent mode: Dynamically adjust the frequency of sonic waves according to the characteristics of the species, specifically:

[0181]

[0182] f0 is the reference repellent frequency, which is set according to the biological auditory sensitivity frequency band, such as f0 = 3kHz for birds, and is used to provide basic repellent capability;

[0183] η is the adjustment coefficient, ranging from [0.2 to 0.8], which is used to control the amplitude of the frequency change with the threat;

[0184] T b is the biological activity threat index;

[0185] T th Threat threshold, the critical value that triggers repulsion, is used to avoid excessive response to low-threat scenarios;

[0186] T max It is the maximum threat value, the upper limit of system design, used to limit the frequency adjustment range;

[0187] Physical protection mode: The insulation protection net is deployed through the coordinated deployment of drone swarms. The mesh density is as follows:

[0188]

[0189] ρ min It is the minimum permissible mesh density, set by insulation requirements, to ensure basic protection performance;

[0190] ρ max It is the maximum permissible mesh density, limited by the material strength, used to prevent structural failure;

[0191] A crit It is the critical protection area, set according to the line voltage level, and is used to dynamically balance the protection strength and ecological permeability;

[0192] A i The activity coverage area of ​​type i organisms;

[0193] Ecological transfer model: deploying intelligent traps to release species-specific pheromones with concentration gradients that meet the following requirements:

[0194]

[0195] is the pheromone concentration gradient, which is used to guide the directional migration of organisms. κ is the diffusion coefficient, which is calibrated by the ambient temperature and humidity to match the biological perception characteristics. is the rate of change of the biological threat index over time, reflecting the dynamic growth or decay of the threat;

[0196] Establish a real-time evaluation model for disposal effects, including:

[0197] The effectiveness indicators of acoustic repellent are as follows:

[0198] E s To quantify the effectiveness of acoustic repellent and the duration of the repellent effect;

[0199] ΔS post is the reduction in biological activity area after treatment;

[0200] ΔS pre is the original value of the biological activity area before disposal;

[0201] β is the decay coefficient, which controls the rate at which effectiveness decreases over time;

[0202] t is time;

[0203] The protection net coverage index is as follows:

[0204] M detected is the set of detected biological activity areas;

[0205] M covered It is the set of areas actually covered by the protection net;

[0206] Dynamically switch the treatment mode combination according to the real-time evaluation model when: When Cc is less than δ, δ is the preset value, and the multi-mode collaborative disposal control is started. The control logic is:

[0207]

[0208] e(t) is the threat deviation, that is, e(t) = Tb(t) - T targete , which is the difference between the current threat and the target threat;

[0209] K p is the proportional gain, ranging from [0.5, 2.0], used to quickly respond to threat changes;

[0210] K d is the differential gain, ranging from [0.01, 0.1], used to suppress overshoot and oscillation;

[0211] K i is the integral gain, eliminating steady-state error;

[0212] K p e(t) is a proportional term that directly responds to the current threat deviation and quickly adjusts the treatment intensity;

[0213] is an integral term, which is used to accumulate historical deviations and eliminate long-term steady-state errors;

[0214] It is a differential term used to predict the threat change trend and suppress overshoot and oscillation;

[0215] Dynamically adjust information such as repellent frequency and protective net density based on the biological threat index to ensure that the intensity of treatment matches the risk and avoid waste of resources or insufficient protection. Combining three modes: acoustic repellent, physical protection, and ecological transfer, it covers short-term expulsion, medium-term isolation, and long-term ecological balance needs, improving the overall protection effectiveness. Through effect evaluation indicators and PID control logic, the combination of treatment modes is dynamically switched to ensure continuous optimization of measures. Species-specific pheromones and sound waves are used for dynamic adjustment to reduce the impact on non-target organisms and the environment.

[0216] A certain transmission line is infested with monkeys, whose climbing behavior may cause a short circuit in the line;

[0217] Monkey activity area A detected i =3m 2 , the preset hazard weight wi = 0.9, the monkey bite risk is high, and the activity duration t i =7 days, time factor k = 0.15;

[0218] Threat Index Calculation:

[0219] Exceeding the threshold T th =8, trigger the disposal plan;

[0220] Multimodal processing execution:

[0221] Acoustic repellent, base frequency f0 = 5kHzf, adjustment coefficient η = 0.6, maximum threat value T max =15;

[0222] Dynamic Frequency Scaling:

[0223] Physical protection net, minimum density ρ min =10 holes / cm 2 , maximum density ρ max =30 holes / cm 2 , critical area A crit =2.5m 2 .

[0224] Mesh density calculation:

[0225] Exceeded the maximum value;

[0226] The pheromone concentration gradient is adjusted according to the rate of threat change. Diffusion coefficient κ = 0.2;

[0227]

[0228] Negative gradients guide monkeys to migrate away from the route;

[0229] Dynamic optimization, monitoring of the effectiveness of sound waves, Assume ΔS post t=2m 2 , ΔS pre =3m 2 , β = 0.1, t = 2 hours:

[0230]

[0231] The protection net coverage Cc = 85%, which does not reach the target value δ = 90%;

[0232] PID control: according to the error e(t) = Tb(t) - T targete =10.8-8=2.8, initiating multi-mode collaboration. Increasing pheromone release and supplementing drone patrols ultimately reduced the biological activity threat index Tb to a safe range.

[0233] The multi-source image data processing in step 1 includes:

[0234] Process the visible light image data, that is, perform equipment deformation detection through the visible light channel and calculate the displacement of the hardware:

[0235] Δd=||SIFT(I t )-SIFT(I t-1 )||2;

[0236] Δd is the displacement of hardware, which represents the deformation and displacement of power line equipment, such as insulators and connecting hardware, at adjacent time points;

[0237] SIFT(·) is a scale-invariant feature transform algorithm used to extract key points and their feature descriptors from images;

[0238] I t is the visible light image at the current time (t), used to capture the real-time status of the device;

[0239] I t-1 is the visible light image at the previous moment (t-1), which serves as the reference for deformation detection;

[0240] Process infrared image data, identify abnormal temperature rise areas through infrared channels, and locate overheating faults at connection points;

[0241] The laser point cloud data is processed, and line sag changes are analyzed through the laser point cloud to detect mechanical stress anomalies.

[0242] The multi-source image data processing in step 1 also involves fusing multi-period image data to construct a 3D degradation model and predict the remaining life of components. The specific process is as follows:

[0243] L r =L0·exp(-λ·∫σ(t)dt);

[0244] L0 is the initial design life of the component, which is used to provide a life prediction benchmark;

[0245] σ(t) is the time-varying stress function, which is used to quantify the impact of the external environment on life;

[0246] λ is the material degradation coefficient, which is calibrated through accelerated aging experiments and is used to reflect the material's fatigue resistance;

[0247] ∫σ(t)dt is the time integral of the time-varying stress, which represents the cumulative external stress to which the component is subjected during its entire life cycle;

[0248] exp is an exponential function used to map the cumulative effect of external stress into a nonlinear relationship of life decay;

[0249] By increasing the intensity of pheromone release and supplementing drone patrols, the threat index Tb of biological activity can ultimately be reduced to a safe range. By combining visible light, infrared, and lidar data, abnormal equipment displacement, temperature rise, and mechanical stress can be located to avoid misjudgments from a single data source. A three-dimensional degradation model is constructed using multi-period data to quantify cumulative component damage and predict remaining lifespan. This supports proactive maintenance decisions and reveals correlations between deformation, temperature rise, and mechanical stress. For example, abnormal sag can lead to localized overheating, improving fault tracing capabilities.

[0250] For example, if the insulators of high-voltage transmission lines age due to long-term mechanical loads and temperature changes, their remaining lifespan needs to be assessed.

[0251] Multi-source data collection and processing:

[0252] Visible light imaging: Using drones to capture images of insulators, the SIFT algorithm is used to detect deformation and displacement.

[0253] Δd=||SIFT(I t )-SIFT(I t-1 )||2 = 12 pixels, exceeding the safety threshold of 5 pixels;

[0254] Infrared imaging: Abnormal temperature at the insulator connection point is detected, such as a local temperature rise of ΔT = 15°C.

[0255] Laser point cloud: Analyze line sag changes and find that mechanical stress exceeds the limit, such as stress value σ = 50MPa;

[0256] 3D degradation model construction and life prediction: Integrate the image data of the past three years, input the initial life L0 = 30 years, and the material degradation coefficient λ = 0.02 years -1 .

[0257] The stress function σ(t) is integrated over time.

[0258] Remaining life calculation:

[0259] L r =L0·exp(-λ·∫σ(t)dt)=30·e -0.02×1200 ≈0.3 years;

[0260] The system determined that the remaining life of the insulator was only about 4 months, far exceeding the critical risk value.

[0261] Triggering preventive maintenance plans: Automatically dispatching drones to replace insulators to avoid sudden breakage and power outages.

[0262] The specific process of step 4 includes:

[0263] Automatically select operating equipment based on fault type: when there is a mechanical fault, a climbing robot is called;

[0264] Insulation inspection drones are activated when there is an electrical fault;

[0265] Activate the sonic repellent device when there is a biologically related fault;

[0266] At the same time, a safe operation path is generated, and the objective function of safe operation path planning is:

[0267]

[0268] R risk The dynamic risk coefficient of the path segment is a 0-1 normalized value that combines the following risk factors:

[0269] Electrical risks: line voltage level, discharge probability, etc.

[0270] Biological risk: Threat index of surrounding biological activities.

[0271] Terrain risk: terrain complexity, such as slope and obstacle density.

[0272] Calculation method: Real-time evaluation through multimodal data to quantify the degree of path danger;

[0273] d i is the Euclidean length of the path segment, which represents the distance the robot arm or drone moves on the path segment. It is used to control the energy consumption of movement. The longer the path, the higher the energy consumption cost.

[0274] T totalThe total time estimate for completing a fault repair operation, including movement time and operation time, such as the time it takes for a climbing robot to replace a component, is used together with the risk cost to form a multi-objective optimization function to avoid inefficiency caused by excessive pursuit of low risk.

[0275] α is the time cost conversion coefficient, which is used to achieve multi-objective optimization;

[0276] Automatically deploy appropriate intelligent operating equipment, such as climbing robots or drones, based on the fault type (mechanical, electrical, or biological), reducing manual intervention delays and the risk of misoperation. A dynamic risk coefficient is generated by integrating electrical, biological, and terrain risks to quantify the degree of path danger.

[0277] By optimizing the path through the objective function, we can minimize the repair time while ensuring safety and avoid overly conservative or risky extreme strategies.

[0278] If the insulators of a high-voltage line in a mountainous area are mechanically damaged due to strong winds, the fault type is a mechanical failure and a climbing robot needs to be called in to replace it.

[0279] The system identifies it as a mechanical failure, automatically dispatches the climbing robot, and loads its operating parameters, such as the load capacity and movement speed of the robotic arm.

[0280] Path planning, risk factors:

[0281] Electrical risk, the line voltage level is 500kV, and the discharge probability is high R elec =0.8;

[0282] Biological risk, bird activity detected in the surrounding area, threat index biological activity threat index Tb = 4, R bio =0.3;

[0283] Terrain risk, the path contains steep slopes and rock obstacles, complexity score R terrain =0.7;

[0284] Dynamic risk factor calculation:

[0285] R risk =0.5·R elec +0.3·R bio +0.2·R terrain =0.5·0.8+0.3·0.3+0.2·0.7=0.63.

[0286] Path segment parameters:

[0287] Path A: length d A =15, risk factor R risk,A =0.63;

[0288] Path B: length dB =20, avoid steep slopes, risk factor R risk,B =0.3.

[0289] Objective function calculation, α = 0.1:

[0290] Total cost of path A: 0.63·15 + 0.1·10 = 10.45, estimated time: 10 minutes;

[0291] Total cost of path B: 0.3·20 + 0.1·15 = 7.5, estimated time: 15 minutes;

[0292] The system chooses Path B, which has a lower total cost and takes slightly longer, but has lower risks.

[0293] The climbing robot was simulated using a three-dimensional model to move along path B. It detected an unmarked loose rock and dynamically adjusted its trajectory to avoid the obstacle. The updated path cost was 7.8, which was still better than path A.

[0294] The process of safe operation path planning also includes:

[0295] Rehearse the disposal process through the digital twin system to detect conflicting solutions;

[0296] Dynamically adjust the robot arm's operating trajectory, specifically:

[0297]

[0298] q0 is the initial angle vector of the robot arm joint, which provides the starting reference position of the motion trajectory and serves as a reference point for trajectory adjustment to ensure that the robot arm starts operation from a known safe state;

[0299] K p is the proportional gain coefficient, which quickly responds to the current trajectory tracking error;

[0300] e(τ) is the trajectory tracking error, that is, the difference between the actual position and the target position, which is used as the basis for feedback control;

[0301] K i is the integral gain coefficient, which is used to eliminate the cumulative effect of historical errors;

[0302] ∫e(τ)dτ is the time integral of the error, which represents the accumulated error from the initial time to the current time t;

[0303] dτ is the differential symbol of time, indicating that the integral variable is time τ;

[0304] By simulating the operation process through the digital twin system, conflicts between the robot arm's motion path and the environment, such as obstacles and live equipment, can be detected in advance to avoid safety accidents during actual execution.

[0305] Dynamically adjust the joint angles of the robotic arm based on proportional-integral (PI) control logic, compensate for positioning errors in real time, and improve operating accuracy in complex environments.

[0306] The system adjusts the trajectory through real-time feedback to adapt to sudden environmental changes, such as equipment deviation caused by strong winds, ensuring operation continuity and reliability.

[0307] The specific contents of step five include:

[0308] Build a fault handling knowledge graph to associate fault characteristics, handling solutions, and effect indicators;

[0309] The decision-making strategy is optimized through reinforcement learning. The decision-making strategy function optimized by reinforcement learning is:

[0310]

[0311] T repair The actual repair time is the time taken from the occurrence of a fault to complete repair, reflecting the handling efficiency. It is calculated based on real-time data such as the movement speed and operation complexity of intelligent operation equipment such as drones and robots;

[0312] T max The maximum allowable repair time is set by power supply reliability standards, such as urban power supply requirements

[0313] T max = 120 minutes, used to standardize time efficiency indicators and ensure that the reward value matches the actual business needs;

[0314] C saved To recover the economic losses, we can calculate the cost by reducing the power outage time and the degree of equipment damage. Example: If the power outage lasts for 1 hour and causes a loss of 500,000 yuan, and the repair time is shortened to 0.5 hours, then C saved =250,000 yuan;

[0315] C total The maximum potential loss amount is set based on the average loss value of similar failures in the same period in history and is used to standardize the economic impact index. For example, if the historical average loss is 1 million yuan, then C total =1 million yuan;

[0316] ω1 and ω2 are time and economic weight coefficients, satisfying ω1+ω2=1, which are used to define the optimization target priority;

[0317] Build a map that correlates fault characteristics, treatment plans, and effect indicators to form a traceable intelligent decision-making system, avoid experience fragmentation, balance repair efficiency and economic impact through reward functions, achieve global optimization, and use reinforcement learning to dynamically adjust strategy weights based on historical repair data to adapt to the needs of different scenarios and improve long-term adaptability.

[0318] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. An intelligent detection method for outdoor power line fault detection, characterized in that: The following steps are involved: Step 1: Multimodal data acquisition and preprocessing: Using drone clusters, distributed fiber optic sensors, lidar, and meteorological monitoring equipment, we simultaneously acquire visible light image data, infrared image data, laser point cloud data, vibration waveforms, temperature distribution, and environmental parameters of power lines. We then process the multi-source image data, i.e., visible light image data, infrared image data, and laser point cloud data. Step 2: Intelligent fault diagnosis: Input the multimodal data preprocessed in step 1 into the multi-task neural network model, output the fault location and type identification results, and obtain the fault type; Step 3: Dynamic decision generation: Based on the analysis of fault type and environmental parameters, a decision plan is generated that includes handling priority, risk level, and resource scheduling; Step 4: Autonomous disposal execution, calling intelligent operation equipment according to the fault type; Step 5: Closed-loop optimization feedback: collect and process the data from steps 1 to 4, update the fault handling knowledge graph, and optimize the decision-making strategy through reinforcement learning.

2. The intelligent detection method for outdoor power line fault detection according to claim 1, characterized in that: The specific process of step 2 includes: Construct spatiotemporal feature tensor: Output failure probability matrix:

3. The intelligent detection method for outdoor power line fault detection according to claim 1, characterized in that: The environmental status analysis in step 3 includes a biological activity risk assessment step, and the specific process is as follows: Detect biological aggregation areas around the line through visible light-thermal imaging fusion; Use voiceprint recognition algorithms to analyze biological species characteristics; Calculate the biological activity threat index Tb.

4. The intelligent detection method for outdoor power line fault detection according to claim 3, characterized in that: After obtaining the biological activity threat index Tb, when the biological activity threat index Tb exceeds the preset threshold, a biological protection disposal plan is generated. The biological protection disposal plan includes: Build a multimodal processing instruction set, including: Sonic repellent mode: Dynamically adjust the frequency of sound waves according to the characteristics of biological species; Physical protection mode: deploy an insulation protection net through the coordinated deployment of drone swarms; Ecological transfer model: deploying smart traps to release species-specific pheromones; Establish a real-time evaluation model for disposal effects, including: The effectiveness indicators of acoustic repellent are as follows: The protection net coverage index is as follows: Dynamically switch the treatment mode combination according to the real-time evaluation model when: And Cc is less than δ, multi-mode collaborative control is activated.

5. The intelligent detection method for outdoor power line fault detection according to claim 1, characterized in that: The multi-source image data processing in step 1 includes: Processing visible light image data, that is, performing equipment deformation detection through the visible light channel and calculating the displacement of hardware; Process infrared image data, identify abnormal temperature rise areas through infrared channels, and locate overheating faults at connection points; The laser point cloud data is processed, and line sag changes are analyzed through the laser point cloud to detect mechanical stress anomalies.

6. The intelligent detection method for outdoor power line fault detection according to claim 5, characterized in that: The multi-source image data processing in step one also includes fusing multi-period image data to construct a three-dimensional degradation model and predict the remaining life of the component.

7. The intelligent detection method for outdoor power line fault detection according to claim 1, characterized in that: The specific process of step 4 includes: Automatically select operating equipment based on fault type: when there is a mechanical fault, a climbing robot is called; Insulation inspection drones are activated when there is an electrical fault; Activate the sonic repellent device when there is a biologically related fault; At the same time, a safe operation path is generated.

8. The intelligent detection method for outdoor power line fault detection according to claim 7, characterized in that: The process of safe operation path planning also includes: Rehearse the disposal process through the digital twin system to detect conflicting solutions; Dynamically adjust the robot arm's operating trajectory.

9. The intelligent detection method for outdoor power line fault detection according to claim 1, characterized in that: The specific contents of step five include: Build a fault handling knowledge graph to associate fault characteristics, handling solutions, and effect indicators; Optimizing decision-making strategies through reinforcement learning.

Images