An intelligent monitoring and identifying system for partial discharge defects of a power transmission line

Abstract

This invention relates to the field of partial discharge defect monitoring technology, specifically disclosing an intelligent monitoring and identification system for partial discharge defects in transmission lines. The system includes an online monitoring network comprising sensor nodes distributed and deployed at preset locations on the transmission line, used to continuously collect time-series data on partial discharge signals, ambient temperature and humidity, and load current of the transmission line, and to perform local edge computing and preliminary early warning. This invention combines a wide-area deployed online monitoring network with on-demand scheduled UAV monitoring units to construct a point-to-surface collaborative perception system, overcoming the shortcomings of existing technologies such as limited coverage of fixed monitoring, isolated and inefficient mobile inspections, and fragmented data sources. It achieves a more comprehensive and flexible perception of the status of transmission lines, enabling operation and maintenance decisions to shift from experience-based periodic general inspections or post-event maintenance to a predictive maintenance model based on accurate status assessment.

Description

Technical Field

[0001] This invention relates to the field of partial discharge defect monitoring technology, specifically to an intelligent monitoring and identification system for partial discharge defects in transmission lines. Background Technology

[0002] As the main artery of power transmission, the insulation health of transmission lines directly affects the safety, stability, and economic operation of the entire power system. Insulation deterioration is one of the main causes of line faults, while partial discharge (PD) is an early and typical sign of local breakdown or surface flashover of insulating materials under the action of a strong electric field. PD activity not only continuously erodes the insulating medium and accelerates its aging process, but may also eventually lead to serious accidents such as insulation breakdown, tripping, or even line breakage, causing huge economic losses and social impacts. Therefore, timely and effective PD monitoring and defect identification of transmission lines are key prerequisites for implementing predictive maintenance, preventing major faults, and ensuring the reliability of the power grid.

[0003] For a long time, partial discharge monitoring of transmission lines has mainly relied on two relatively independent technical means: one is to install fixed monitoring devices on key towers or line segments. These devices can continuously collect partial discharge signals at specific locations, but due to high deployment costs and limited coverage, it is difficult to achieve full-line status awareness. The other is to conduct manual inspections using handheld detectors or manned helicopters. This method is flexible but inefficient, greatly affected by the environment and human factors, and it is difficult to obtain quantitative and continuous monitoring data, posing a risk of missed detections and misjudgments. In recent years, although there have been attempts to use drones equipped with detection equipment for inspections, they are usually used as independent data collection tools, lacking data and command coordination with fixed monitoring networks, failing to form an integrated monitoring capability. Moreover, most of them can only provide alarms for the presence of partial discharge or simple signal strength, unable to reveal the evolution trend of defects, quantify their severity, or assess the remaining safe life of components. This makes operation and maintenance decisions lack accurate data support, often resulting in a crude mode of periodic general inspections or post-event maintenance, which is not only costly to operate and maintain but also difficult to effectively prevent sudden failures. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent monitoring and identification system for partial discharge defects in transmission lines, solving the following technical problems:

[0005] The question is how to achieve deep fusion of multi-source monitoring data and deep intelligent assessment and early warning of partial discharge defects.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A smart monitoring and identification system for partial discharge defects in transmission lines, the system comprising:

[0008] The online monitoring network includes sensor nodes distributed at preset locations on the transmission lines, which are used to continuously collect time-series data on partial discharge signals, ambient temperature and humidity, and load current of the transmission lines, and to perform local edge computing and preliminary early warning.

[0009] The UAV monitoring unit includes the UAV body, an airborne multi-dimensional sensor module and a ground control station. The airborne multi-dimensional sensor module integrates at least an ultra-high frequency sensor, an ultrasonic sensor and an ultraviolet imager, and is used to conduct close-range inspections of designated power transmission line areas or components according to mission instructions, and collect high spatial resolution partial discharge diagnostic data.

[0010] The data fusion and analysis platform is used to receive and fuse multi-source heterogeneous monitoring data from online monitoring networks and UAV monitoring units, quantify the evolution stage of partial discharge defects in transmission line components and predict the remaining safe operating time, and generate risk assessment and maintenance decision reports.

[0011] The human-computer interaction terminal is used to provide maintenance personnel with risk assessment and maintenance decision reports, real-time early warning signals, and to receive feedback information from maintenance personnel on on-site handling.

[0012] Furthermore, the data fusion and analysis platform includes:

[0013] The model management module is used to create and maintain a dynamically updated component health assessment model for each monitored preset location. The core of this model is a state vector that updates over time. This is used to quantify the overall state and changing trend of partial discharge defects in the component;

[0014] The data fusion module is used to preprocess the input time-series data and diagnostic data, and drive the state vector of the corresponding component. The update process includes spatiotemporal alignment, feature extraction, and noise reduction. The spatiotemporal alignment uses a timestamp synchronization algorithm based on the geographical coordinates of transmission line towers. The feature extraction uses a joint extraction algorithm based on the time-frequency domain features of partial discharge signals. The noise reduction uses a composite algorithm combining wavelet threshold denoising and adaptive Kalman filtering.

[0015] The defect analysis module includes a stage quantification model and a remaining time prediction model, used to analyze the updated state vector. Calculate the defect evolution stage and remaining safe operating time of the component;

[0016] The decision generation module is used to automatically generate a decision report containing risk level, handling suggestions and expected maintenance time window based on the defect evolution stage and remaining safe operating time, combined with component identity and power grid operating conditions.

[0017] Furthermore, the state vector It is a multidimensional feature vector, and its expression is:

[0018]

[0019] in, The j-th dynamic feature value is dynamically calculated from the time-series data and diagnostic data preprocessed by the data fusion module. The state feature value includes at least the moving average value of the partial discharge signal amplitude. The linear growth slope of the signal amplitude over a preset time period Statistical skewness calculated based on phase-resolved partial discharge spectrum and the characteristic frequency of ultrasonic signals ; These are static characteristic values ​​related to component attributes and the environment. The static characteristic values ​​include at least the component type code, historical defect type label, and average operating load coefficient.

[0020] Furthermore, the process of calculating the defect evolution stage of the component using the stage quantification model includes:

[0021] By using the state vector The dynamic feature values ​​are compared with preset multi-level thresholds and logically judged to divide the defect evolution process into multiple stages;

[0022] The aforementioned stages include at least an nascent stage, a stable stage, and a deterioration stage;

[0023] The conditions for determining a period of deterioration are defined by the following set of inequalities:

[0024]

[0025] in, This is a preset rise stage threshold for the moving average value corresponding to the amplitude of the partial discharge signal. This is a preset threshold for the rising phase, corresponding to the linear growth slope of the signal amplitude. This is the preset threshold for the rising stage corresponding to statistical skewness.

[0026] Furthermore, the process by which the remaining time prediction model calculates the remaining safe operating time of the component includes:

[0027]

[0028] The remaining safe operating time of the component is obtained through analysis and calculation using the above formula. ,

[0029] in, The current overall condition score of the component is obtained through analysis and calculation using the following formula;

[0030]

[0031] in, , The current state vector The i-th eigenvalue, For the i-th eigenvalue The first weighting coefficient, To map eigenvalues ​​to a normalized function with uniform dimensions, Preset safety thresholds for component shutdown procedures. Based on the historical comprehensive condition score sequence of this component The historical average degradation rate was calculated. The current degradation rate is calculated based on the recent changes in the component's overall condition score. , This is the second weighting coefficient. This represents the current defect evolution stage output by the stage quantification model. This is the stage deterioration coefficient, and its value is positively correlated with the severity of the defect evolution stage.

[0032] Furthermore, the first weighting coefficient The model is dynamically trained and generated by the model management module based on machine learning algorithms, and core features are given higher weights.

[0033] The normalization function The parameterized mapping function is expressed as follows:

[0034]

[0035] in, This is the slope adjustment parameter. This is the offset parameter.

[0036] Furthermore, the data fusion and analysis platform also includes a feedback optimization module, used to receive on-site handling results input from the human-computer interaction terminal, wherein the feedback data serves as a labeled sample;

[0037] The feedback optimization module performs the following optimization process:

[0038] S1. According to preset rules, the parameters to be optimized are divided into two groups with a sequential order, including the multi-level threshold of the stage quantization model and the safety state threshold of the remaining time prediction model. For the first priority optimization group, the stage deterioration coefficient is... The mapping relationship and the second weight coefficient , Slope adjustment parameters and offset parameters It is the second optimization group;

[0039] S2. Under the condition of fixed data fusion module parameters and second optimization group parameters, iteratively update the parameters of the first optimization group until its output error converges to the first preset tolerance range.

[0040] S3. While keeping the optimization results of the first optimization group parameters basically stable, start iterative updates of the second optimization group parameters until the output error converges to the second preset tolerance range.

[0041] S4. After completing the parameter optimization of the stage quantization model and the remaining time prediction, make small-scale adaptive fine-tuning of the noise reduction and feature extraction parameters in the data fusion module based on the optimized model.

[0042] S5. In any iteration of step S2 or step S3, if the error is not significantly reduced after a preset number of iterations, or if the error shows a diverging trend, the pause mechanism is triggered to stop the current stage of optimization and optionally roll back to the parameter state before the start of this optimization.

[0043] Furthermore, the data fusion and analysis platform also includes a task scheduling module, which is used to automatically generate a refined inspection task sheet for drones based on the early warning signals from the online monitoring network, the predicted risks output by the defect analysis module, or the preset inspection plan.

[0044] The generation of the UAV-based refined inspection task sheet is based on at least one of the following triggering logics:

[0045] Response-based triggering: When the online monitoring data of a component is processed by local edge computing, any feature value exceeds its adaptive warning threshold;

[0046] Predictive triggering: When the remaining time prediction model outputs the remaining safe runtime. Less than the preset warning time threshold ;

[0047] Phase verification trigger: When the phase quantification model determines that a defect in a component has entered a new evolution phase.

[0048] The beneficial effects of this invention are:

[0049] (1) This invention combines a wide-area online monitoring network with on-demand UAV monitoring units to construct a point-to-surface collaborative perception system, overcoming the shortcomings of limited fixed monitoring coverage, isolated and inefficient mobile inspections, and fragmented data sources in the prior art. It achieves a more comprehensive and flexible perception of the status of transmission lines. Furthermore, the system further conducts in-depth mining and intelligent analysis of multi-source heterogeneous data through a data fusion analysis platform. This not only enables the detection of partial discharge defects but also quantifies their evolution stages and predicts the remaining safe operating time. As a result, the output is upgraded from a simple risk warning to a multi-dimensional in-depth assessment report that includes the nature of the risk, its development trend, and its urgency. This allows maintenance decisions to shift from a periodic general inspection or post-event maintenance model that relies on experience to a predictive maintenance model based on accurate status assessment. This effectively solves the problems of high maintenance costs and difficulty in effectively preventing sudden failures in the prior art.

[0050] (2) This invention uses a set of stage quantification models based on clear rules and thresholds to clearly divide the development of defects into stages such as budding, stabilization and deterioration, thus realizing the qualitative classification of risk. Furthermore, through a remaining time prediction model that integrates the current state, historical trends, recent changes and stage correction factors, it realizes the quantitative prediction of risk life. This method not only defines the stage in which the defect is located, but also predicts the remaining service life of the component. It also ensures the adaptability and accuracy of the evaluation model through dynamic weights and parameterized functions, thereby improving the operation and maintenance decision from fuzzy judgment based on experience to precise guidance based on data and models. Attached Figure Description

[0051] The invention will now be further described with reference to the accompanying drawings.

[0052] Figure 1 This is a schematic block diagram of an intelligent monitoring and identification system for partial discharge defects in power transmission lines proposed in this invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Please see Figure 1 As shown, in one embodiment, an intelligent monitoring and identification system for partial discharge defects in transmission lines is provided, the system comprising:

[0055] The online monitoring network includes sensor nodes distributed at preset locations on the transmission line, which are used to continuously collect time-series data of partial discharge signals, ambient temperature and humidity, and load current of the transmission line. The raw data is preprocessed locally (such as filtering and preliminary feature extraction) and preliminary anomaly warning is given through the edge computing unit built into the node, forming an all-weather perception layer covering the backbone nodes of the line. The preset locations are key locations determined by risk assessment, such as easily polluted areas, important crossing points, and old line sections.

[0056] The UAV monitoring unit includes the UAV itself, an airborne multi-dimensional sensor module, and a ground control station. The airborne multi-dimensional sensor module integrates multiple complementary detection methods such as UHF sensors, ultrasonic sensors, and ultraviolet imagers. It can autonomously or under control fly to the airspace above specific components such as designated towers, insulators, or hardware according to the task instructions issued by the data fusion and analysis platform, and perform close-range, multi-angle fine scanning to collect partial discharge diagnostic data with higher spatial resolution and better signal-to-noise ratio, which can be used to supplement and verify the information of fixed monitoring points.

[0057] The data fusion and analysis platform, deployed on a cloud server or local data center, is used to receive and fuse multi-source heterogeneous monitoring data from online monitoring networks and UAV monitoring units, quantify the evolution stage of partial discharge defects in transmission line components and predict the remaining safe operating time, and generate risk assessment and maintenance decision reports. The multi-source heterogeneous monitoring data includes wide-area time-series data streams from online monitoring networks and precise inspection data from UAV monitoring units.

[0058] The human-computer interaction terminal is used to provide maintenance personnel with risk assessment and maintenance decision reports, real-time early warning signals, and to receive feedback information from maintenance personnel on on-site handling.

[0059] Through the above technical solution, this embodiment provides an intelligent monitoring and identification system for partial discharge defects in transmission lines. This system combines a widely deployed online monitoring network with on-demand dispatchable UAV monitoring units to construct a point-to-surface collaborative perception system. This overcomes the shortcomings of existing technologies, such as limited coverage of fixed monitoring, isolated and inefficient mobile inspections, and fragmented data sources. It achieves a more comprehensive and flexible perception of the transmission line status. Furthermore, the system uses a data fusion analysis platform to deeply mine and intelligently analyze multi-source heterogeneous data. This not only detects partial discharge defects but also quantifies their evolution stages and predicts remaining safe operating time. The output is upgraded from a simple risk warning to a multi-dimensional in-depth assessment report that includes the nature of the risk, its development trend, and its urgency. This allows maintenance decisions to shift from experience-based periodic inspections or post-event maintenance to a predictive maintenance model based on accurate status assessment. This effectively solves the problems of high maintenance costs and difficulty in effectively preventing sudden failures in existing technologies.

[0060] In one embodiment, the data fusion analysis platform includes:

[0061] The model management module is used to create and maintain a dynamically updated component health assessment model for each pre-defined location of the monitored tower, insulator, or fitting. The core of this model is a state vector that updates over time. This is used to quantify the overall state and changing trend of partial discharge defects in the component;

[0062] The data fusion module is used to preprocess the input time-series data and diagnostic data, and drive the state vector of the corresponding component. The update process includes spatiotemporal alignment, feature extraction, and noise reduction. The spatiotemporal alignment employs a timestamp synchronization algorithm based on the precise geographical coordinates of transmission line towers to address the issue of consistent time and spatial references for data collected from different sensors at different times, ensuring data comparability. Feature extraction utilizes a joint extraction algorithm based on the time-frequency domain features of partial discharge signals. This algorithm analyzes not only the signal's time-domain features (such as amplitude and rise time) but also its frequency-domain features (such as dominant frequency and spectral distribution), thereby comprehensively capturing the mode information of the partial discharge signal and providing rich feature dimensions for state assessment. Noise reduction employs a composite algorithm combining wavelet thresholding and adaptive Kalman filtering. Specifically, wavelet transform is first used to remove instantaneous pulse interference and high-frequency electromagnetic noise from the partial discharge signal, and then an adaptive Kalman filter is used to smooth environmental background noise and sensor background noise, significantly improving the signal-to-noise ratio.

[0063] The defect analysis module includes a stage quantization model and a remaining time prediction model, wherein the stage quantization model is based on the updated state vector. By comparing dynamic feature values ​​with multi-level thresholds trained from expert knowledge bases or historical data and making logical decisions, the continuous evolution process of defects is discretized and staged to achieve qualitative cognition. The remaining time prediction model is based on the updated state vector. By integrating historical degradation trends with the current state, the remaining safe operating time of components before failure can be quantitatively predicted, thereby achieving quantitative risk warning.

[0064] The decision generation module is used to automatically generate a decision report containing risk level, handling suggestions and expected maintenance time window based on the defect evolution stage and remaining safe operating time, combined with the inherent attributes of the component (such as model and voltage level) and the current power grid operating conditions (such as load and weather).

[0065] The state vector It is a multidimensional feature vector, and its expression is:

[0066]

[0067] in, The j-th dynamic feature value is dynamically calculated from the time-series data and diagnostic data preprocessed by the data fusion module, and its state feature value changes with time t. This state feature value includes at least the moving average of the partial discharge signal amplitude. The linear growth slope of the signal amplitude over a preset time period Statistical skewness calculated based on phase-resolved partial discharge spectrum and the characteristic frequency of ultrasonic signals The moving average value of the partial discharge signal amplitude It can be obtained by averaging the amplitude of partial discharge pulses over a past time window (e.g., 1 minute). This reflects the average intensity of the discharge within the time window and is a fundamental indicator for assessing the severity of the discharge. The linear growth slope mentioned... This can be achieved by analyzing a preset time period (e.g., the past hour). The slope obtained by linear fitting of the sequence reveals a key trend indicator of whether the discharge intensity tends to stabilize, grows slowly, or accelerates. A positive value and a larger value indicate faster deterioration. This statistical skewness... The phase-resolved spectrum of high-quality partial discharge pulse signals collected by UAVs within the power frequency cycle can be statistically calculated. This spectrum characterizes the concentration or pattern of discharge on the voltage phase. Different types of defects (such as internal discharge and surface discharge) have different skewness characteristics, which are important bases for defect pattern recognition. The characteristic frequency of the ultrasonic signal... By performing spectrum analysis on the audio signals collected by ultrasonic sensors, the frequency component with the highest energy can be extracted. Since defects such as mechanical loosening and crack discharge will generate ultrasonic waves of specific frequencies, the ultrasonic signals collected by UAVs at close range and in a directional manner are less affected by environmental wind noise and can more accurately extract this characteristic frequency to determine the physical type of the defect. These are static characteristic values ​​related to component attributes and the environment, which are relatively stable or change slowly. The static characteristic values ​​include at least the component type code, historical defect type label, and average operating load factor. The component type code is an identifier for insulators, surge arresters, or cable terminals, etc., used to match the corresponding evaluation criteria and thresholds. The historical defect type label can record the types of defects that have occurred in the past for risk tracking and pattern continuity judgment. The average operating load factor can reflect the long-term electrical stress level that the component has endured. A high load factor may accelerate insulation aging and is a correction factor for assessing life reduction.

[0068] Through the above technical solution, this embodiment provides a method for in-depth quantitative assessment and intelligent decision-making of partial discharge defects in transmission line components. The method constructs a dynamic state vector as a unified quantitative framework and updates the vector through a professional data fusion process. This enables the defect analysis module to perform accurate stage quantification and remaining life prediction based on this vector. As a result, partial discharge monitoring is elevated from the traditional alarm based on the presence or absence of a signal to a deeper level of understanding of the status, trend, and remaining life. Ultimately, it automatically generates an actionable decision report, thereby realizing a fundamental shift in operation and maintenance decision-making from experience-driven to data-driven, and from passive response to proactive prediction.

[0069] In one embodiment, the stage quantization model is used to discretize the continuous defect evolution process into multiple stages with distinct characteristics, and the process of calculating the defect evolution stages of a component includes:

[0070] By using the state vector The system compares and logically judges key dynamic feature values ​​with preset multi-level thresholds to divide the defect evolution process into multiple stages. The multi-level thresholds include at least baseline thresholds, warning thresholds, and escalation thresholds corresponding to different stages. These thresholds can be determined based on statistical analysis of a large amount of historical normal data, defect development process data, and failure case data, combined with equipment manufacturer technical specifications, industry standards, and domain expert experience. For example, the baseline threshold can be determined by statistically analyzing the feature value distribution of similar equipment during normal operation; the warning threshold and escalation threshold can be determined by analyzing the change nodes of feature values ​​during the defect's development from observable to faulty. This provides a hierarchical and quantifiable scale for the objective and standardized assessment of defect status. It can transform vague descriptions of state deterioration into clear judgments of exceeding a specific threshold, enabling the system to automatically and consistently classify component status into different risk stages, providing a decision-making basis for subsequent differentiated warnings and precise operation and maintenance.

[0071] The multiple phases include at least the nascent stage (characteristic value exceeds the baseline threshold but is below the warning threshold), the stable stage (characteristic value fluctuates around the warning threshold), and the deterioration stage (characteristic value reaches or exceeds the escalation threshold).

[0072] The conditions for determining a period of deterioration are defined by the following set of inequalities:

[0073]

[0074] in, The preset threshold for the rising stage of the moving average of the partial discharge signal amplitude can be determined based on a large amount of historical data statistics and simulation analysis. It represents the critical value for the average intensity of partial discharge of this type of component to enter the accelerated degradation zone, and is a hard indicator for judging whether the discharge intensity meets the standard. The preset threshold for the linear growth slope of the corresponding signal amplitude can be determined by analyzing cases where defects transition from the stable period to the deterioration period and statistically analyzing the lower limit of their amplitude growth rate. It is a key indicator for judging whether the discharge activity shows an accelerating trend. The preset threshold for the rising stage corresponding to the statistical skewness can be set according to the typical phase distribution characteristics of different defect modes (such as corona and surface discharge). It is an indicator for judging whether the discharge mode has undergone a qualitative change (such as changing from uniform corona to dangerous local jets).

[0075] The process by which the remaining time prediction model calculates the remaining safe operating time of a component includes:

[0076]

[0077] The remaining safe operating time of the component is obtained through analysis and calculation using the above formula. ,

[0078] in, The current overall condition score of the component is given, which directly reflects the component's current overall health level. It is obtained through analysis and calculation using the following formula;

[0079]

[0080] in, , The current state vector The i-th eigenvalue, For the i-th eigenvalue The first weight coefficient is dynamically generated by the model management module through periodic calls to machine learning algorithms such as random forests, based on historical data and diagnostic results, and assigns higher weights to core features. To map eigenvalues ​​to a normalized function with uniform dimensions, The safety state thresholds for component shutdown are defined by operation and maintenance standards and safety procedures, or derived by reverse engineering through failure cases. When the value falls below this threshold, the component is considered to need to be taken out of service immediately; this is the endpoint of the lifespan prediction. Based on the historical comprehensive condition score sequence of this component The calculated historical average degradation rate reflects the long-term, inherent aging or degradation trend of the component. The current degradation rate is calculated based on the recent (e.g., within 24 hours) changes in the component's overall condition score, reflecting the component's recent and potentially accelerating degradation dynamics. , This is the second weighting coefficient, which can be preset or adaptively adjusted based on the model validation results, and is used to balance long-term patterns. Recent mutations The proportion of contribution in the prediction This represents the current defect evolution stage output by the stage quantification model. The deterioration coefficient is the factor for the stage, based on The values ​​are obtained through a preset mapping table or function, and their values ​​are positively correlated with the severity of the defect evolution stage, for example, the nascent stage. =0.8, stable period =1.0, deterioration period =1.5, used to dynamically amplify or reduce the degradation rate in the prediction formula, so that the prediction results can better reflect the urgency of risks at different stages.

[0081] The normalization function The parameterized mapping function is expressed as follows:

[0082]

[0083] in, This is a slope adjustment parameter, whose initial value comes from a knowledge base or expert experience. It can be optimized through feedback and controls the steepness of the function curve. It is used to match the sensitivity of different feature values ​​as the defect develops. The offset parameter, whose initial value comes from a knowledge base or expert experience, can be optimized through feedback. It determines the location of the function's center point, i.e., the eigenvalue. At what value is it mapped to an intermediate health score (e.g., 0.5)?

[0084] Through the above technical solution, this embodiment provides an intelligent hierarchical early warning and lifetime quantitative prediction method for partial discharge defects in transmission lines. The method uses a stage quantitative model based on clear rules and thresholds to clearly divide the development of defects into stages such as nascent, stable, and deteriorating, thus achieving qualitative risk classification. Furthermore, through a remaining time prediction model that integrates current status, historical trends, recent changes, and stage correction factors, it achieves quantitative prediction of risk lifetime. This method not only defines the stage of the defect but also predicts the remaining service life of the component. Dynamic weights and parameterized functions ensure the adaptability and accuracy of the evaluation model, thereby elevating operation and maintenance decisions from fuzzy judgments based on experience to precise guidance based on data and models.

[0085] In one embodiment, the data fusion analysis platform further includes a feedback optimization module, used to receive on-site handling result feedback input from the human-computer interaction terminal, wherein the feedback data serves as a labeled sample, that is, it contains pairing information between the input data and the final confirmation result;

[0086] The feedback optimization module performs the following optimization process:

[0087] S1. According to preset rules, the parameters to be optimized are divided into two groups with a sequential order, including the multi-level threshold of the stage quantization model and the safety state threshold of the remaining time prediction model. For the first priority optimization group, the stage deterioration coefficient is... The mapping relationship and the second weight coefficient , Slope adjustment parameters and offset parameters It is the second optimization group;

[0088] S2. Under the condition of fixed data fusion module parameters and second optimization group parameters, the parameters of the first optimization group are iteratively updated. The labeled samples (known real defect stages or whether shutdown should be performed) are used as supervision signals. The goal is to minimize the loss function between the current system output and the real label (such as cross-entropy loss for stage judgment and mean square error for threshold deviation). Numerical optimization methods such as gradient descent, grid search or Bayesian optimization are used to search for the optimal solution in the feasible space defined by the parameters. The iteration continues until the output error converges to the first preset tolerance range (such as stage judgment accuracy > 95%).

[0089] S3. While maintaining the basic stability of the optimization results of the first optimization group parameters, start the iterative update of the second optimization group parameters. The core of this stage of optimization is that it does not rely on externally set absolute truth labels, but rather on the complete state evolution sequence of each defective component being addressed. and its final disposal time It is viewed as a self-supervised signal source containing information on degradation patterns and urgency. The optimization process aims to drive the model parameters to converge towards a principle: so that the model's output for the same component sequence exhibits a consistent pattern in the time dimension with the common sense of physical degradation and operational logic. Specifically, this is manifested in the following ways:

[0090] Trend consistency principle: the overall state score output by the model It should show a non-increasing trend over time, reflecting the basic physical constraint that health will not recover without cause;

[0091] Predictive consistency principle: the remaining safe running time predicted by the model. It should match the trend of its output health score; that is, the faster the health score declines, the more significantly the predicted remaining time should be shortened.

[0092] Alignment principle for handling boundaries: The model should be aligned with the boundary near the handling time. The risk assessment output (such as phases and remaining time) should correspond in urgency to the actual response actions taken, for example, in We should not make overly optimistic long-term predictions about the remaining time.

[0093] To achieve the above principles, the optimization process aims to maximize the consistency between the model output sequence and the aforementioned inherent laws. A corresponding measurable optimization objective is constructed (for example, a loss function that encourages temporal consistency). Numerical optimization methods are used to search and iteratively update the parameters of the second optimization group. The iteration continues until the consistency measure between the model output sequence and the inherent laws converges to the second preset tolerance range.

[0094] S4. After completing the parameter optimization of the stage quantization model and the remaining time prediction, make small-scale adaptive fine-tuning of the denoising and feature extraction parameters in the data fusion module based on the optimized model. For example, adjust the threshold of wavelet denoising to better retain the signal frequency bands that are most important for the current model to judge, or optimize the parameters of the feature extraction algorithm to enhance the sensitivity to the main defect patterns that have been identified.

[0095] S5. In any iteration of step S2 or step S3, a pause and rollback mechanism is introduced to ensure the stability of the optimization process and prevent system performance degradation caused by data noise, labeling conflicts or optimization algorithm problems. If the error is not significantly reduced after a preset number of iterations (e.g., 10 times) or the error shows a diverging trend, the pause mechanism is triggered to stop the current stage of optimization. It is also possible to automatically roll back the parameter group currently being optimized to the stable state before the start of this optimization cycle. This ensures that failed optimization attempts will not damage the existing reliability of the system.

[0096] Through the above technical solutions, this embodiment provides a progressive and robust parameter self-optimization method. The method ensures the priority calibration and stability of the core system safety criteria through rigorous parameter grouping and optimization stage division. Based on the learning paradigm of relative degradation trend and urgency of handling, it solves the problem of missing absolute truth labels in traditional methods, enabling the model to learn more fundamental physical laws of degradation. Through front-end fine-tuning and robustness guarantee mechanisms, it realizes full-link collaborative adaptation from signal processing to intelligent analysis, thereby evolving the system from a static configuration tool into an intelligent diagnostic platform with continuous self-calibration and self-evolution capabilities. This directly and effectively solves the core defects of existing technologies, such as fixed system parameters and inability to adapt to dynamic and complex environments.

[0097] In one embodiment, the data fusion analysis platform further includes a task scheduling module, which is used to automatically generate a refined inspection task list for UAVs based on the early warning signals from the online monitoring network, the predicted risks output by the defect analysis module, or the preset inspection plan. The task list includes at least the precise geographical coordinates of the target component (such as tower number, phase, and hardware location), the recommended UAV approach path, the sensor types that need to be activated (such as ultraviolet, ultrasonic, or UHF), and the data specifications to be collected.

[0098] The generation of the UAV-based refined inspection task sheet is based on at least one of the following triggering logics:

[0099] Response-based triggering: When the online monitoring data of a component is processed by local edge computing, any key feature value extracted (such as the moving average of the partial discharge signal amplitude) is triggered. If the device exceeds its adaptive warning threshold, the adaptive warning threshold is not a fixed value, but is obtained based on the historical normal operation data of the component (such as taking the historical average value plus 3 times the standard deviation), and can be dynamically fine-tuned according to factors such as season and load. It is a primary and sensitive benchmark for judging whether the equipment status deviates from the historical normal baseline, and is used to capture sudden or early weak anomalies.

[0100] Predictive triggering: When the remaining time prediction model outputs the remaining safe runtime. Less than the preset warning time threshold The preset warning time threshold It can be set comprehensively based on the safety procedures for power grid operation and maintenance, the maintenance resource allocation cycle, and risk tolerance. For example, if planned maintenance usually requires two weeks of preparation, then... It can be set to 30 days, which can transform quantitative life prediction into a specific action schedule to achieve predictive maintenance. When the predicted life enters this warning window, it means that a confirmatory inspection needs to be arranged to provide the final on-site basis for subsequent possible maintenance decisions.

[0101] Phase-based verification trigger: When the phase quantification model determines that a defect in a component has entered a new evolution phase (e.g., from the "emergence stage" to the "stable stage," or from the "stable development stage" to the "deterioration stage"), the new evolution phase is determined by the phase quantification model based on the state vector. The comparison results with multi-level thresholds show that the transition of the defect stage indicates a qualitative change in its potential risk nature or development speed. This triggering logic ensures that drones are dispatched for refined diagnosis as soon as a critical change in the risk nature occurs, in order to obtain the most realistic on-site data for updating the model and confirming the risk level.

[0102] Through the above technical solution, this embodiment provides an intelligent task scheduling and precise inspection triggering method for monitoring partial discharge defects in transmission lines. The method integrates three layers of triggering logic: responsive (based on real-time anomalies), predictive (based on remaining lifetime), and phased verification (based on state transitions). This constructs a multi-dimensional, proactive UAV task scheduling system, changing the traditional pattern of periodic inspections or passive responses. It ensures that each UAV deployment has a clear and high-level diagnostic purpose, thereby greatly improving the targeting of inspections and the utilization efficiency of maintenance resources, and realizing the transformation from blind inspections to precise detection.

[0103] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent monitoring and identification system for partial discharge defects in transmission lines, characterized in that, The system includes: The online monitoring network includes sensor nodes distributed at preset locations on the transmission lines, which are used to continuously collect time-series data on partial discharge signals, ambient temperature and humidity, and load current of the transmission lines, and to perform local edge computing and preliminary early warning. The UAV monitoring unit includes the UAV body, an airborne multi-dimensional sensor module and a ground control station. The airborne multi-dimensional sensor module integrates at least an ultra-high frequency sensor, an ultrasonic sensor and an ultraviolet imager, and is used to conduct close-range inspections of designated power transmission line areas or components according to mission instructions, and collect high spatial resolution partial discharge diagnostic data. The data fusion and analysis platform is used to receive and fuse multi-source heterogeneous monitoring data from online monitoring networks and UAV monitoring units, quantify the evolution stage of partial discharge defects in transmission line components and predict the remaining safe operating time, and generate risk assessment and maintenance decision reports. The human-computer interaction terminal is used to provide maintenance personnel with risk assessment and maintenance decision reports, real-time early warning signals, and to receive feedback information from maintenance personnel on on-site handling.

2. The intelligent monitoring and identification system for partial discharge defects in transmission lines according to claim 1, characterized in that, The data fusion and analysis platform includes: The model management module is used to create and maintain a dynamically updated component health assessment model for each monitored preset location. The core of this model is a state vector that updates over time. This is used to quantify the overall state and changing trend of partial discharge defects in the component; The data fusion module is used to preprocess the input time-series data and diagnostic data, and drive the state vector of the corresponding component. The update process includes spatiotemporal alignment, feature extraction, and noise reduction. The spatiotemporal alignment uses a timestamp synchronization algorithm based on the geographical coordinates of transmission line towers. The feature extraction uses a joint extraction algorithm based on the time-frequency domain features of partial discharge signals. The noise reduction uses a composite algorithm combining wavelet threshold denoising and adaptive Kalman filtering. The defect analysis module includes a stage quantification model and a remaining time prediction model, used to analyze the updated state vector. Calculate the defect evolution stage and remaining safe operating time of the component; The decision generation module is used to automatically generate a decision report containing risk level, handling suggestions and expected maintenance time window based on the defect evolution stage and remaining safe operating time, combined with component identity and power grid operating conditions.

3. The intelligent monitoring and identification system for partial discharge defects in transmission lines according to claim 2, characterized in that, The state vector It is a multidimensional feature vector, and its expression is: in, The j-th dynamic feature value is dynamically calculated from the time-series data and diagnostic data preprocessed by the data fusion module. The state feature value includes at least the moving average value of the partial discharge signal amplitude. The linear growth slope of the signal amplitude over a preset time period Statistical skewness calculated based on phase-resolved partial discharge spectrum and the characteristic frequency of ultrasonic signals ; These are static characteristic values ​​related to component attributes and the environment. The static characteristic values ​​include at least the component type code, historical defect type label, and average operating load coefficient.

4. The intelligent monitoring and identification system for partial discharge defects in transmission lines according to claim 3, characterized in that, The process of calculating the defect evolution stage of a component using the stage quantification model includes: By using the state vector The dynamic feature values ​​are compared with preset multi-level thresholds and logically judged to divide the defect evolution process into multiple stages; The aforementioned stages include at least an nascent stage, a stable stage, and a deterioration stage; The conditions for determining a period of deterioration are defined by the following set of inequalities: in, This is a preset rise stage threshold for the moving average value corresponding to the amplitude of the partial discharge signal. This is a preset threshold for the rising phase, corresponding to the linear growth slope of the signal amplitude. This is the preset threshold for the rising stage corresponding to statistical skewness.

5. The intelligent monitoring and identification system for partial discharge defects in transmission lines according to claim 4, characterized in that, The process by which the remaining time prediction model calculates the remaining safe operating time of a component includes: The remaining safe operating time of the component is obtained through analysis and calculation using the above formula. , in, The current overall condition score of the component is obtained through analysis and calculation using the following formula; in, , The current state vector The i-th eigenvalue, For the i-th eigenvalue The first weighting coefficient, To map eigenvalues ​​to a normalized function with uniform dimensions, Preset safety thresholds for component shutdown procedures. Based on the historical comprehensive condition score sequence of this component The historical average degradation rate was calculated. The current degradation rate is calculated based on the recent changes in the component's overall condition score. , This is the second weighting coefficient. This represents the current defect evolution stage output by the stage quantification model. This is the stage deterioration coefficient, and its value is positively correlated with the severity of the defect evolution stage.

6. The intelligent monitoring and identification system for partial discharge defects in transmission lines according to claim 5, characterized in that, The first weighting coefficient The model is dynamically trained and generated by the model management module based on machine learning algorithms, and core features are given higher weights. The normalization function The parameterized mapping function is expressed as follows: in, This is the slope adjustment parameter. This is the offset parameter.

7. The intelligent monitoring and identification system for partial discharge defects in transmission lines according to claim 6, characterized in that, The data fusion and analysis platform also includes a feedback optimization module, which is used to receive on-site handling results input from the human-computer interaction terminal, and the feedback data serves as a labeled sample. The feedback optimization module performs the following optimization process: S1. According to preset rules, the parameters to be optimized are divided into two groups with a sequential order, including the multi-level threshold of the stage quantization model and the safety state threshold of the remaining time prediction model. For the first priority optimization group, the stage deterioration coefficient is... Mapping relationship, second weight coefficient , Slope adjustment parameters and offset parameters It is the second optimization group; S2. Under the condition of fixed data fusion module parameters and second optimization group parameters, iteratively update the parameters of the first optimization group until its output error converges to the first preset tolerance range. S3. While keeping the optimization results of the first optimization group parameters basically stable, start iterative updates of the second optimization group parameters until the output error converges to the second preset tolerance range. S4. After completing the parameter optimization of the stage quantization model and the remaining time prediction, make small-scale adaptive fine-tuning of the noise reduction and feature extraction parameters in the data fusion module based on the optimized model. S5. In any iteration of step S2 or step S3, if the error is not significantly reduced after a preset number of iterations, or if the error shows a diverging trend, the pause mechanism is triggered to stop the current stage of optimization and optionally roll back to the parameter state before the start of this optimization.

8. The intelligent monitoring and identification system for partial discharge defects in transmission lines according to claim 7, characterized in that, The data fusion and analysis platform also includes a task scheduling module, which is used to automatically generate a refined inspection task sheet for drones based on the early warning signals from the online monitoring network, the predicted risks output by the defect analysis module, or the preset inspection plan. The generation of the UAV-based refined inspection task sheet is based on at least one of the following triggering logics: Response-based triggering: When the online monitoring data of a component is processed by local edge computing, any feature value exceeds its adaptive warning threshold; Predictive triggering: When the remaining time prediction model outputs the remaining safe runtime. Less than the preset warning time threshold ; Phase verification trigger: When the phase quantification model determines that a defect in a component has entered a new evolution phase.

Images