A method and apparatus for detecting early arc faults in power plant electrical equipment.
By combining the bipolar matching degree of the polarity spatiotemporal neighborhood with the physical law of the directional diffusion of electric arcs, false events are eliminated and electric arc characteristics are enhanced. This solves the problems of low sensitivity and high false alarm rate in existing electric arc detection technologies, and realizes real-time accurate detection and graded early warning of early electric arc faults in power plant electrical equipment.
Patent Information
- Application Number
- CN202611253889.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-08-18
Smart Images

Figure CN122778271B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of arc warning technology, and specifically relates to a method and device for detecting early arc faults in electrical equipment of power plants. Background Technology
[0002] Insulation degradation of electrical equipment in power plants is a core cause of equipment failure, and early weak arcs (also known as early arcs) are the most direct and earliest indicator of insulation degradation. Among them, early arcs caused by insulation degradation of electrical equipment under normal pressure often last for microseconds and are characterized by weak light intensity and low discharge energy. If they cannot be identified in time, they will quickly develop into continuous arcs, eventually leading to insulation breakdown, equipment burnout, unplanned unit shutdown, and even major safety accidents such as fires or explosions. Therefore, there is a strong and rigid demand for accurate and interference-resistant detection of early arcs in electrical equipment.
[0003] Currently, early arc fault detection in electrical equipment mainly includes traditional electrical quantity detection methods, traditional visible light imaging detection methods, and arc detection technology based on event cameras. Among these, arc detection technology based on event cameras utilizes an event camera (Dynamic Vision Sensor, DVS) to sense changes in scene light intensity with a microsecond-level time resolution, outputting only an asynchronous event stream of brightness changes. It has advantages such as low bandwidth, low power consumption, and high dynamic range. Compared with traditional electrical quantity detection methods and traditional visible light imaging detection methods, arc detection technology based on event cameras is the optimal technical path for detecting transient arc signals.
[0004] However, existing arc detection technologies using event cameras generally suffer from low detection sensitivity, high false alarm rates, and poor real-time performance. Specifically, firstly, power plant scenarios are characterized by specific noises such as direct / obstructed sunlight, inspection lighting, metal equipment reflections, and strong electromagnetic interference, generating a large number of false events. The event stream denoising algorithms in existing event camera arc detection technologies cannot distinguish between environmental false events and arc events. They either fail to denoise thoroughly, resulting in a high false alarm rate, or they filter out sparse early arcs as noise, easily leading to missed detections. Secondly, early arcs caused by insulation degradation have extremely sparse event streams and very low signal-to-noise ratios. Existing technologies do not incorporate targeted feature enhancement based on the physical evolution of arcs, making them easily submerged by background noise, resulting in high missed detection rates and failing to achieve true early warning. Furthermore, existing technologies are mostly purely data-driven general algorithms that do not address the physical nature of arc discharge, resulting in poor generalization in complex industrial environments. They also mostly only achieve a binary determination of the presence or absence of an arc, without designing a tiered early warning mechanism based on the needs of power plant operation and maintenance, making them unsuitable for direct engineering applications. Summary of the Invention
[0005] To address the technical problems existing in the prior art, this invention provides a method and apparatus for early arc fault detection in power plant electrical equipment, in order to solve the technical problems of low detection sensitivity, high false alarm rate and poor real-time performance that are common in existing event camera arc detection technologies.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a method for detecting early arc faults in power plant electrical equipment, comprising: Obtain the raw event set of the device under test; wherein, the raw event set is the set of raw events output by the event camera, which contains several events arranged in chronological order; Based on the bipolar matching degree of the polarity spatiotemporal neighborhood of the event, ambient light unipolar pseudo-events in the original event set are removed to obtain a preliminary set of valid events; By combining the spatiotemporal continuity of the electric arc, the spatiotemporal joint credibility weight of each event in the initial screening effective event set is calculated, and invalid noise events are removed to obtain an event set with credibility weights. Based on the physical laws of the directional diffusion of electric arcs, and combined with the spatiotemporal joint credibility weight of each event in the event set with credibility weight, early electric arc feature directional enhancement is performed on the events in the event set with credibility weight to obtain an event set with enhanced feature amplitude. By using the sliding window feature extraction technique, multidimensional feature extraction is performed on the event set with enhanced feature amplitude to obtain the multidimensional feature vector of arc fault; Based on the multidimensional feature vector of arc faults, the early arc fault detection results of the device under test are obtained through multi-feature fuzzy fusion processing.
[0007] Furthermore, based on the bipolarity matching degree of the event's spatiotemporal neighborhood, the process of eliminating ambient light unipolar pseudo-events from the original event set to obtain the initial screening of valid event sets is as follows: Calculate the bipolarity matching degree of the spatiotemporal neighborhood of each event in the original event set; The bipolarity matching degree of the spatiotemporal neighborhood of each event is compared with the preset matching degree threshold. The ambient light unipolar pseudo-events in the original event set are removed to obtain the initial set of effective events. The process of calculating the bipolar matching degree of the polarity spatiotemporal neighborhood for each event in the original event set is as follows:
[0008] in, For the first The bipolar matching degree of the spatiotemporal neighborhood of an event; For the first Spatial coordinates of an event Centered on, spatial neighborhood radius Within, and the timestamp is within the time window. Inside, with the The total number of events with opposite polarities; For the first The trigger timestamp of each event; The time neighborhood is half the width; For the first Spatial coordinates of an event Centered on, spatial neighborhood radius Within, and the timestamp is within the time window. Within, the total number of events of all polarities; To prevent the elimination of the zero smoothing factor.
[0009] Furthermore, considering the spatiotemporal continuity of the electric arc, the spatiotemporal joint confidence weight of each event in the initial screening of valid events is calculated, and invalid noise events are removed to obtain the event set with confidence weights. The process is as follows: Based on the bipolar matching degree of the spatiotemporal neighborhood of each event in the initial screening effective event set, the spatiotemporal joint credibility weight of each event in the initial screening effective event set is calculated using the Gaussian credibility weighted model constrained by the spatiotemporal neighborhood. The spatiotemporal joint credibility weight of each event is compared with the preset credibility weight threshold, and invalid noise events are eliminated to obtain a set of events with credibility weights. The Gaussian credibility-weighted model with spatiotemporal neighborhood constraints is as follows:
[0010] in, For the first Spatiotemporal joint credibility weights of events; For the first The spatiotemporal neighborhood set of an event; For the first The spatiotemporal neighborhood set of an event The first The sequence number of each event; For the first The bipolar matching degree of the spatiotemporal neighborhood of an event; For the first The spatiotemporal neighborhood set of an event Total number of events within; To prevent the removal of the zero smoothing factor; The standard deviation of the Gaussian kernel in space; The standard deviation of the Gaussian kernel over time; For the first The ordinate of the pixel coordinate system for each event; For the first The ordinate of the pixel coordinate system for each event; For the first The trigger timestamp of each event; For the first The trigger timestamp of each event; For the first The x-coordinate of the pixel coordinate system for each event; For the first The vertical coordinate of the pixel coordinate system for each event.
[0011] Furthermore, the process of early arc feature-oriented enhancement for events in the event set with confidence weights is as follows:
[0012] in, For the first The amplitude of the enhanced features of each event; For the first Spatiotemporal joint credibility weights of events; The feature enhancement coefficient; For the first The set of spatiotemporal evolution neighborhoods of an event; For the first The spatiotemporal evolution neighborhood of an event The first The sequence number of each event; For the first Spatiotemporal joint credibility weights of events; For from the first The event to the 1 The unit motion direction vector of each event; is the typical unit direction vector of arc diffusion in the device under test; To prevent the elimination of the zero smoothing factor.
[0013] Furthermore, by utilizing the sliding window feature extraction technique, multidimensional feature extraction is performed on the event set with enhanced feature amplitude to obtain the multidimensional feature vector of the arc fault, as follows:
[0014] in, This represents the multidimensional feature vector of an arc fault. This represents the average enhancement amplitude characteristic; The characteristic is the proportion of high-amplitude events; It is characterized by spatiotemporal dispersion; It is characterized by polarity balance; This is a characteristic of the maximum transient event burst rate; For event sets with enhanced feature amplitude Total number of events; For a set of events with enhanced feature amplitude; For the first The amplitude of the enhanced features of each event; For the sliding window to satisfy The number of events; To prevent the removal of the zero smoothing factor; The standard deviation of all event space coordinates within the sliding window; The standard deviation of all event timestamps within the sliding window; Polarity within the sliding window The number of events; Polarity within the sliding window The number of events; For time Within the time sub-window of the center's statistics, the following conditions are met. The number of events; High amplitude threshold; The width of the time sub-window is used for statistics.
[0015] Furthermore, based on the multi-dimensional feature vector of the arc fault, the process of outputting the early arc fault detection result of the device under test through multi-feature fuzzy fusion processing is as follows: Calculate the membership degree of each feature in the multidimensional feature vector of arc faults; The comprehensive membership degree of arc fault is calculated based on the membership degree of each feature in the multidimensional feature vector of arc fault. Based on the comprehensive membership degree of arc faults, and according to the preset arc fault judgment logic, the early arc fault detection results of the equipment under test are obtained.
[0016] The present invention also provides an early arc fault detection system for power plant electrical equipment, used to implement the aforementioned early arc fault detection method for power plant electrical equipment, comprising: The event acquisition module is used to acquire the raw event set of the device under test; wherein, the raw event set is the raw event set output by the event camera, which contains several events arranged in chronological order; The event screening module is used to remove ambient light unipolar pseudo-events from the original event set based on the bipolarity matching degree of the event's spatiotemporal neighborhood, thereby obtaining a preliminary set of valid events. The noise event filtering module is used to combine the spatiotemporal continuity of the electric arc to calculate the spatiotemporal joint credibility weight of each event in the initial screening of valid events, and to remove invalid noise events to obtain an event set with credibility weights. The feature orientation enhancement module is used to perform early arc feature orientation enhancement on events in the event set with confidence weights based on the physical law of directional diffusion of electric arcs, and combined with the spatiotemporal joint confidence weight of each event in the event set with confidence weights, to obtain an event set with enhanced feature amplitude. The feature extraction module is used to extract multidimensional features from an event set with enhanced feature amplitude using sliding window feature extraction technology, thereby obtaining a multidimensional feature vector of arc fault. The fault determination module is used to output the early arc fault detection results of the equipment under test by processing the multi-dimensional feature vector of the arc fault through multi-feature fuzzy fusion.
[0017] The present invention also provides an electronic device, comprising: A processor is used to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, performs the method for detecting early arc faults in power plant electrical equipment.
[0018] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for early arc fault detection of power plant electrical equipment.
[0019] The present invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the method for early arc fault detection of power plant electrical equipment.
[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: The method for detecting early arc faults in power plant electrical equipment provided by this invention sequentially performs initial screening using bipolar matching degree in a polarity spatiotemporal neighborhood, fine screening using spatiotemporal joint credibility weight, feature enhancement based on the directional diffusion physical laws of arcs, sliding window feature extraction, and multi-feature fuzzy fusion processing. This achieves real-time and accurate detection of early arcs in complex power plant environments, offering advantages such as strong resistance to industrial environmental interference, high detection sensitivity, low false alarm rate, strong real-time performance, and good engineering adaptability. Specifically, based on the physical nature of simultaneous generation of positive and negative ions and paired occurrence of bipolar events during real arc discharge, the method identifies and eliminates unipolar pseudo-events caused by unidirectional changes in ambient light through bipolar matching degree in a polarity spatiotemporal neighborhood, effectively avoiding the risk of missed detection due to false filtering of early arcs. Based on the physical law that arc plasma has continuous evolution characteristics in the spatiotemporal dimension, a spatiotemporal joint credibility weight is assigned to each event, achieving the detection of arcs in real-time and accurate detection of arcs in complex environments. The differentiated distinction between scattered noise events and continuous arc events allows sparse arc events to achieve higher confidence in subsequent processing. Based on the physical mechanism of arc discharge energy directional diffusion along the plasma, early arc features are enhanced in a directional manner, overcoming the limitations of pure data-driven algorithms that ignore the physical nature of the arc and cannot directionally amplify extremely weak signals, thus effectively improving the signal-to-noise ratio of microsecond-level arc events. Multi-dimensional physical features of arc faults are extracted through a sliding window, and multi-feature fuzzy fusion is used to achieve hierarchical output of multi-level faults to adapt to the gradient response requirements of power plant operation and maintenance scenarios for fault severity. This invention is not a simple transplantation of general event stream processing algorithms, but rather deeply embeds the physical evolution law of arc discharge into each processing step, forming a closed-loop detection with physical mechanism constraints and multi-level progressive processing, thereby simultaneously meeting the requirements of high sensitivity, low false alarm rate, and strong real-time arc detection of power plant electrical equipment.
[0021] The early arc fault detection system, electronic equipment, computer-readable storage medium, and computer program product for power plant electrical equipment provided by this invention possess all the advantages of the aforementioned early arc fault detection methods for power plant electrical equipment. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 A flowchart of the early arc fault detection method for power plant electrical equipment provided in Example 1; Figure 2 This is a structural block diagram of the early arc fault detection system for power plant electrical equipment provided in Example 2; Figure 3 This is a structural block diagram of the electronic device provided in Example 3. Detailed Implementation
[0024] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0025] Before describing the specific embodiments of this application, some of the technical terms involved in the embodiments of this application are explained as follows: Early weak arc, also known as early arc, refers to the weak arc phenomenon in the early stage of insulation defects in electrical equipment, which is low in energy, unstable, and close to corona or partial discharge. As a precursor to fault, it is easily masked by noise and difficult to detect by conventional protection.
[0026] An event camera is an asynchronous, event-driven bionic vision sensor that independently detects changes in light intensity at the pixel level and outputs only visual change information. It features high temporal resolution, ultra-wide dynamic range, low latency, and low data redundancy.
[0027] This invention provides a method for detecting early arc faults in electrical equipment of power plants, comprising the following steps: Step 100: Obtain the original event set of the device to be tested; wherein, the original event set is the original event set output by the event camera, which contains several events arranged in chronological order.
[0028] Step 200: Based on the bipolarity matching degree of the event's spatiotemporal neighborhood, eliminate ambient light unipolar pseudo-events in the original event set to obtain a preliminary set of valid events.
[0029] Step 300: Combining the spatiotemporal continuity of the electric arc, calculate the spatiotemporal joint credibility weight of each event in the initial screening of valid events, and remove invalid noise events to obtain an event set with credibility weights.
[0030] Step 400: Based on the physical laws of the directional diffusion of electric arcs, and combined with the spatiotemporal joint credibility weight of each event in the event set with credibility weight, perform early electric arc feature directional enhancement on the events in the event set with credibility weight to obtain an event set with enhanced feature amplitude.
[0031] Step 500: Using the sliding window feature extraction technique, perform multi-dimensional feature extraction on the event set with enhanced feature amplitude to obtain the multi-dimensional feature vector of arc fault.
[0032] Step 600: Based on the multi-dimensional feature vector of the arc fault, the early arc fault detection result of the device under test is obtained through multi-feature fuzzy fusion processing.
[0033] In the above implementation, the original event set output by the event camera is first acquired. Based on the bipolar matching degree of the event's spatiotemporal neighborhood, unipolar pseudo-events caused by direct sunlight, equipment reflection, and strong electromagnetic interference in the power plant scene are accurately eliminated, completing the initial screening and noise reduction. Then, combining the spatiotemporal continuity of the electric arc, the spatiotemporal joint credibility weight of each event in the initial screening effective event set is calculated to further eliminate invalid noise events, forming a two-level anti-interference noise reduction mechanism. This effectively solves the problem of incomplete noise reduction or false filtering of early electric arcs in existing technologies, and can significantly reduce the false alarm rate and false negative rate. On this basis, combining the physical law of the directional diffusion of the electric arc, the spatiotemporal joint credibility weight of the event is used to perform directional enhancement of early electric arc features, effectively highlighting the microsecond-level early electric arc feature signals and avoiding them being submerged by background noise, thereby achieving true... This technology provides early fault warning in a true sense. It employs sliding window feature extraction and multi-feature fuzzy fusion processing, eliminating the need for complex deep learning models and enabling real-time operation on embedded hardware. This ensures a latency of less than 50ms from event acquisition to alarm output, meeting the real-time requirements of online monitoring. Furthermore, the entire detection process is designed based on the physical nature of arc discharge, rather than simple data fitting, resulting in strong generalization and applicability to various electrical equipment in power plants. The detection results can also meet the needs of power plant operation and maintenance management, providing reliable support for practical engineering implementation. This effectively solves the problems of poor generalization and inability to directly adapt to engineering applications in existing technologies, fundamentally improving the accuracy, anti-interference capabilities, real-time performance, and engineering practicality of early arc detection for power plant electrical equipment. It can effectively prevent equipment failures and major safety accidents caused by insulation degradation.
[0034] The following specific embodiments further explain the method for detecting early arc faults in power plant electrical equipment provided by the present invention: Example 1 As attached Figure 1 As shown, this embodiment 1 provides a method for detecting early arc faults in power plant electrical equipment, including the following steps: Step 1: Obtain the raw event set of the device to be tested; wherein, the raw event set is the raw event set output by the event camera, which contains several events arranged in chronological order.
[0035] Specifically, the implementation process is as follows: The asynchronous event stream acquired in real time by the event camera is read, and the asynchronous event stream is windowed according to a sliding window. The original event set within each sliding window is used as the original event set of the device under test. The original event set of the device under test is denoted as:
[0036]
[0037] in, The original event set of the device to be tested; For the first event in the original event set One event; A unique sequence number for the event; The total number of events within the sliding window or the original event set; For the first The x-coordinate of the pixel coordinate system for each event; For the first The ordinate of the pixel coordinate system for each event; For the first The trigger timestamp of each event; For the first The polarity of an event, +1 indicates that the corresponding pixel brightness is increased, and -1 indicates that the corresponding pixel brightness is decreased.
[0038] It should be noted that the event camera is pre-positioned at a preset location within the safety protection area of the device to be tested, and the field of view calibration, device installation posture calibration, and typical arc diffusion direction calibration of the event camera are completed in advance; the device to be tested is, for example, a generator slip ring, a transformer bushing, or a high-voltage switchgear; optionally, a fixed sliding time window is used, with a width of 1000μs, which can be adaptively adjusted according to the on-site scene to balance alarm delay requirements and event statistical effectiveness.
[0039] Step 2: Based on the bipolarity matching degree of the event's spatiotemporal neighborhood, eliminate ambient light unipolar pseudo-events in the original event set to obtain the initial set of valid events.
[0040] Explanatoryly, ambient light unipolar pseudo-events refer to pseudo-events caused by environmental interference in power plant scenarios, such as direct / obstructed sunlight, inspection lighting, metal equipment reflection, or strong electromagnetic interference. In power plant scenarios, ambient light unipolar pseudo-events account for more than 60%. In step 2, based on the physical differences between electric arcs and environmental interference, ambient light unipolar pseudo-events in the original event set are screened out, which can remove more than 90% of noise events while retaining electric arc events completely. The output of step 2 includes the initially screened effective event set and the polarity spatiotemporal neighborhood bipolar matching degree of each event in the initially screened effective event set, providing a basic quantitative basis for subsequent spatiotemporal joint credibility weighting and feature enhancement.
[0041] Specifically, the implementation process is as follows: Step 21: Calculate the bipolarity matching degree of the spatiotemporal neighborhood of each event in the original event set; the formula for calculating the bipolarity matching degree of the spatiotemporal neighborhood of each event in the original event set is as follows:
[0042] in, For the first The bipolarity matching degree of the polarity spatiotemporal neighborhood of an event. The larger the value of the bipolarity matching degree of the polarity spatiotemporal neighborhood, the higher the probability that the event belongs to an electric arc event. For the first Spatial coordinates of an event Centered on, spatial neighborhood radius Within, and the timestamp is within the time window. Inside, with the The total number of events with opposite polarities; The time neighborhood is half the width; For the first Spatial coordinates of an event Centered on, spatial neighborhood radius Within, and the timestamp is within the time window. Within, the total number of events of all polarities; To prevent the use of a zero smoothing factor, and to avoid calculation errors where the denominator of the formula is zero when there are no events in the neighborhood.
[0043] The derivation principle of the formula is explained as follows: In this embodiment 1, considering the ambient light interference in the power plant scenario, based on the polarity output mechanism of the event camera and combined with the instantaneous light radiation characteristics of the arc plasma, the bipolarity matching degree of the polarity spatiotemporal neighborhood of each event in the original event set is calculated. Among these, ambient light unipolar pseudo-events generated by sunlight, inspection lights, and reflections in the power plant scenario are characterized by large area, same polarity, and synchronous triggering. Their neighborhoods contain almost only events of a single polarity. Even in the early stages of an arc, plasma emission will simultaneously trigger positive and negative bipolar events in the same spatiotemporal neighborhood, which is the essential physical difference between the two. Based on this, a formula for calculating the bipolarity matching degree of the polarity spatiotemporal neighborhood of each event in the original event set is constructed. The bipolarity matching degree of the polarity spatiotemporal neighborhood of each event is calculated, allowing subsequent use of the bipolarity matching degree as an indicator. By statistically analyzing the proportion of opposite polarity events in the spatiotemporal neighborhood of a single event, the probability of an event belonging to an arc event is quantified, thereby achieving accurate initial screening of ambient light unipolar pseudo-events.
[0044] It should be noted that the polarity output mechanism of the event camera is an inherent physical property of the event camera and will not be elaborated here; the instantaneous light radiation characteristics of the arc plasma refer to the rapid rise and fall of light intensity in a local space during the excitation and recombination of the plasma during normal pressure arc discharge. Within a microsecond timescale, positive and negative bipolar events will be triggered simultaneously in the same spatiotemporal neighborhood; spatial neighborhood radius The value is based on the following: the initial luminous region diameter of the early electric arc is approximately 3-8 pixels, and the spatial neighborhood radius... The value of needs to completely cover the neighborhood range of the arc event; optionally, the spatial neighborhood radius Take 5 pixels; temporal neighborhood half width The value is based on the fact that the change in arc light radiation is on the order of microseconds and the half-width of the time neighborhood is... The value must fully cover the triggering range of a single arc pulse; optionally, the time neighborhood half-width Take 50 μs.
[0045] Step 22: Compare the bipolarity matching degree of the spatiotemporal neighborhood of each event with a preset matching degree threshold, and remove ambient light unipolar pseudo-events from the original event set to obtain the initial set of valid events. The principle for removing ambient light unipolar pseudo-events from the original event set is as follows:
[0046] in, This serves as an initial screening of valid event sets; This is the preset matching threshold.
[0047] Optionally, since the bipolar matching degree of the polarity spatiotemporal neighborhood of an electric arc event is typically greater than or equal to 0.2, while the bipolar matching degree of the polarity spatiotemporal neighborhood of a unipolar pseudo-event of ambient light in a power plant scenario is typically less than 0.05; therefore, a preset matching degree threshold is set. Setting the value to 0.2 allows for effective screening of electric arc events.
[0048] Compared to existing technologies, current event stream denoising algorithms only denoise based on the spatiotemporal density and isolation of events, without utilizing the bipolar physical characteristics of electric arcs. Therefore, they cannot specifically eliminate ambient light unipolar pseudo-events in power plant scenarios. In this embodiment 1, the bipolarity matching degree of the polarity spatiotemporal neighborhood is used as an indicator. By statistically analyzing the proportion of opposite polarity events within the spatiotemporal neighborhood of a single event, the probability of an event belonging to an electric arc event is quantified, achieving accurate initial screening of ambient light unipolar pseudo-events. Specifically, when specifically eliminating ambient light unipolar pseudo-events in power plant scenarios, the screening rate can reach over 90%, significantly reducing the data processing volume in subsequent steps and ensuring the real-time performance of the method. Furthermore, it can completely retain electric arc events with bipolar physical characteristics, preventing false filtering even of sparse electric arc events, thus reducing the false negative rate at its source.
[0049] Step 3: Combining the spatiotemporal continuity of the electric arc, calculate the spatiotemporal joint credibility weight of each event in the initial screening of valid events, and remove invalid noise events to obtain an event set with credibility weights.
[0050] Explanatoryly, in this embodiment 1, for the residual thermal noise, dark current noise, and a small number of residual ambient light unipolar pseudo-events in the initial screening effective event set, based on the spatiotemporal continuity characteristics of the electric arc, the spatiotemporal joint credibility weight of each event is quantified to further suppress random interference noise; the output of step 3 includes an event set with credibility weights and the spatiotemporal joint credibility weight of each event, realizing the transfer from the bipolar matching degree of the polarity spatiotemporal neighborhood to the spatiotemporal joint credibility weight, providing a weight basis for subsequent feature enhancement, and ensuring information continuity.
[0051] Specifically, the implementation process is as follows: Step 31: Based on the bipolar matching degree of the spatiotemporal neighborhood of each event in the initial screening of valid events, calculate the spatiotemporal joint credibility weight of each event in the initial screening of valid events using a Gaussian credibility weighted model constrained by spatiotemporal neighborhood; wherein, the Gaussian credibility weighted model constrained by spatiotemporal neighborhood is as follows:
[0052] in, For the first The spatiotemporal joint credibility weight of an event is determined by the fact that the larger the spatiotemporal joint credibility weight, the higher the credibility of the event as an electric arc event. For the first The spatiotemporal neighborhood set of an event; For the first The spatiotemporal neighborhood set of an event The first The sequence number of each event; For the first The bipolar matching degree of the spatiotemporal neighborhood of an event; For the first The spatiotemporal neighborhood set of an event Total number of events within; The standard deviation of the Gaussian kernel in space; The standard deviation of the Gaussian kernel over time; For the first The ordinate of the pixel coordinate system for each event; For the first The ordinate of the pixel coordinate system for each event; For the first The trigger timestamp of each event.
[0053] Explanatoryly, the low-matching interference remaining in the initial screening of valid events, such as thermal noise and dark current noise from the event camera itself, and a small number of ambient light unipolar pseudo-events; the thermal noise and dark current noise from the event camera itself manifest as isolated, spatiotemporally uncorrelated random events; the small number of ambient light unipolar pseudo-events have extremely weak spatiotemporal correlation; however, electric arcs have clear spatial and temporal continuity. Spatial continuity refers to the characteristic that the electric arc spreads along the discharge channel and is continuously triggered by adjacent pixels, while temporal continuity refers to the characteristic that the electric arc is continuously triggered within microseconds; classic Gaussian filtering only considers spatial or temporal distance constraints, cannot distinguish between valid events and random noise, and cannot fuse the physical characteristics of electric arcs.
[0054] In this embodiment 1, by fusing the spatiotemporal joint credibility weights of each event in the initial screening of valid events, a spatial and temporal dual Gaussian kernel constraint is constructed. Simultaneously considering the spatial and temporal continuity of the electric arc, a Gaussian credibility weighted model with spatiotemporal neighborhood constraints is formed based on the Gaussian kernel neighborhood filtering algorithm in digital image processing and combined with the spatiotemporal correlation model in event camera processing. By fusing the spatiotemporal joint credibility weights of each event in the initial screening of valid events, a higher base weight can be provided for high-matching electric arc events, further strengthening the contribution of electric arc events and suppressing residual low-matching interference. By constructing a spatial and temporal dual Gaussian kernel constraint, simultaneously considering the spatial and temporal continuity of the electric arc, the spatiotemporal continuity characteristics of the electric arc are accurately matched. Unlike single-dimensional filtering algorithms, the Gaussian credibility weighted model with spatiotemporal neighborhood constraints can output normalized credibility weights to quantify the credibility of each event belonging to an electric arc event, providing a unified quantitative benchmark for subsequent feature enhancement.
[0055] Step 32: Compare the spatiotemporal joint credibility weight of each event with a preset credibility weight threshold, and remove invalid noise events to obtain a set of events with credibility weights; wherein, the set of events with credibility weights is represented as follows:
[0056] in, This is a set of events with credibility weights.
[0057] In this embodiment 1, the preset confidence weight threshold is set to 0.05; the residual low-matching interference is unrelated events in the neighborhood, and the bipolar matching degree of the polar spatiotemporal neighborhood of unrelated events is close to zero; therefore, the preset confidence weight threshold of 0.05 can filter out random interference noise with a bipolar matching degree of less than 0.05 in the polar spatiotemporal neighborhood; at the same time, it can assign high confidence weights to events that conform to the spatiotemporal continuity characteristics of electric arcs, quantify the effectiveness of each event, and achieve the initial enhancement of electric arc events.
[0058] Step 4: Based on the physical laws of the directional diffusion of electric arcs, and combined with the spatiotemporal joint credibility weight of each event in the event set with credibility weight, perform early electric arc feature directional enhancement on the events in the event set with credibility weight to obtain an event set with enhanced feature amplitude.
[0059] Explanatoryly, considering the characteristics of early electric arcs, such as high sparsity, low feature amplitude, and susceptibility to background noise, this paper uses the directional diffusion physics of electric arc plasma to perform directional feature enhancement on events that conform to the characteristics of electric arc evolution. This further widens the feature gap between electric arc events and noise events, thereby reducing the false negative rate from the root. The output of step 4 includes a set of events with enhanced feature amplitude and the enhanced feature amplitude of each event, providing a basis for subsequent multidimensional feature extraction.
[0060] Specifically, the process of early arc feature-oriented enhancement for events in a set of events with confidence weights is as follows:
[0061]
[0062] in, For the first The amplitude of the enhanced feature of an event is the value of the enhanced feature. The larger the value of the enhanced feature amplitude, the more significant the characteristic of the event as an electric arc event is. The feature enhancement coefficient; For the first The set of spatiotemporal evolution neighborhoods of an event; For the first The spatiotemporal evolution neighborhood of an event The first The sequence number of each event; For the first Spatiotemporal joint credibility weights of events; For from the first The event to the 1 The unit motion direction vector of each event; is the typical unit direction vector of arc diffusion in the device under test; For the first The x-coordinate of the pixel coordinate system for each event; For the first The vertical coordinate of the pixel coordinate system for each event.
[0063] The event set with enhanced feature magnitude is represented as follows:
[0064] in, For the event set with enhanced feature amplitude.
[0065] Explanatoryly, even after weighting the spatiotemporal joint credibility weights of each event in the credibility weighted event set, arc events still suffer from high sparsity, low feature amplitude, and are easily submerged by background. Direct feature extraction can lead to serious missed detections. The core physical law of arc discharge is that the arc spreads directionally along the insulation degradation path or along the direction of electric field strength, and there is a clear consistency in the direction of motion between adjacent arc events. In contrast, the direction of motion of noise events is completely random and has no consistency. Existing feature enhancement algorithms are mostly general amplitude amplification, which is prone to the problem of simultaneously amplifying both signal and noise, and cannot specifically enhance the features of early arcs.
[0066] In this embodiment 1, based on the spatiotemporal joint credibility weight of each event in the event set with credibility weight, and combined with the spatiotemporal evolution dynamics of arc plasma, i.e., the physical law of arc directional diffusion, and the matching filter enhancement principle in signal processing, early arc feature directional enhancement is performed. Specifically, the spatiotemporal evolution dynamics of arc plasma are as follows: under normal pressure, the arc caused by the insulation degradation of electrical equipment will exhibit directional spatial diffusion characteristics along the insulation defect path or along the direction of electric field strength, and the diffusion direction is highly correlated with the direction of electric field strength. The matching filter enhancement principle in signal processing, for example, is to perform directional enhancement of the target signal by matching with the prior features of the target signal, and suppress non-target signals.
[0067] In this embodiment 1, based on the spatiotemporal joint credibility weight of each event in the event set with credibility weight, it can ensure that high-credibility arc events have basic weights and that low-credibility noise events are not amplified. Secondly, by introducing the physical law of directional diffusion of arcs, the matching degree between the motion direction of neighborhood events and the typical diffusion direction of arcs is statistically analyzed. Only events that conform to the physical law of directional diffusion of arcs are enhanced in a targeted manner, while noise events with random directions are not enhanced or are even suppressed, thus solving the core defect of general enhancement algorithms. In addition, the output enhancement feature amplitude can significantly improve the signal-to-noise ratio of arc events, providing high signal-to-noise ratio feature data for subsequent feature quantization and recognition.
[0068] Optionally, feature enhancement coefficient The value is set to 1.5, based on the principle of enhancing event characteristics that conform to the physical laws of directional arc diffusion by 50%-150% without amplifying noise, balancing enhancement effect and anti-interference capability. This is the characteristic enhancement coefficient. The value can be adjusted adaptively based on the actual situation; The spatiotemporal evolution neighborhood of an event The definition of is: with the first Spatial coordinates of an event Centered on, spatial neighborhood radius Within, and the timestamp is within the time window. All events within; optionally, time neighborhood. The value of 50 μs was chosen based on: the historical diffusion trajectory of the covered arc event, conforming to microsecond-level evolution characteristics; and the spatial neighborhood radius. The value is set to 8 pixels, based on the principle of covering the spatial diffusion range of the electric arc and fully capturing the diffusion trajectory.
[0069] It should be noted that in this embodiment 1, based on the physical law of the directional diffusion of electric arcs, early electric arc features are enhanced in a targeted manner for events in the event set with confidence weights, which greatly improves the feature significance of early electric arcs and solves the defects of early electric arcs being easily submerged by noise and having a high false negative rate. Secondly, only events that conform to the physical law of the directional diffusion of electric arcs are enhanced to avoid amplifying random dry noise and further widen the feature gap between electric arc events and noise events, providing high signal-to-noise ratio feature data for subsequent feature quantization and identification.
[0070] Step 5: Using the sliding window feature extraction technique, perform multi-dimensional feature extraction on the event set with enhanced feature amplitude to obtain the multi-dimensional feature vector of arc fault.
[0071] Explanatoryly, in step 5, the event set with enhanced feature amplitude is transformed into a quantized feature vector that is highly sensitive to electric arc and insensitive to random interference noise, realizing the transformation from discrete event data to structured feature data, and providing a comprehensive and reliable quantitative basis for the final fault determination; specifically, based on the sliding window feature extraction technology, multidimensional feature extraction formula is used to extract multidimensional features from the event set with enhanced feature amplitude to obtain a multidimensional feature vector of electric arc fault; optionally, the multidimensional feature vector of electric arc fault is a five-dimensional electric arc fault vector.
[0072] Specifically, the multidimensional feature extraction formula is as follows:
[0073]
[0074] in, This represents the multidimensional feature vector of an arc fault. The average enhancement amplitude characteristic within the window; The characteristic is the proportion of high-amplitude events; It is characterized by spatiotemporal dispersion; It is characterized by polarity balance; This is a characteristic of the maximum transient event burst rate; For event sets with enhanced feature amplitude Total number of events; For a set of events with enhanced feature amplitude; For the sliding window to satisfy The number of events; The standard deviation of all event space coordinates within the sliding window; The standard deviation of all event timestamps within the sliding window; Polarity within the sliding window The number of events; Polarity within the sliding window The number of events; For time Within the time sub-window of the center's statistics, the following conditions are met. The number of events; High amplitude threshold; For statistical purposes, use a time sub-window; This represents the average of the vertical coordinates of all events within the sliding window; This represents the average of the vertical axes of all events within the sliding window.
[0075] Explanatoryly, after performing early arc feature-oriented enhancement on events in a set of events with confidence weights, the feature amplitude of individual events is enhanced. However, to achieve accurate identification of arc events, the spatiotemporal information of all events within the sliding window needs to be transformed into quantitative features that are highly sensitive to early arcs. However, existing event flow feature extraction methods mostly use general statistical features such as event rate and spatiotemporal density. Secondly, existing technologies do not design features specific to the characteristics of arcs and are not sensitive enough to early arcs. In addition, existing technologies have a single feature dimension and cannot comprehensively distinguish between arc events and noise events.
[0076] In this embodiment 1, for the event set with enhanced feature amplitude, a multi-dimensional feature extraction formula is established in combination with the fault characteristics of early arcs, so that all features are deeply integrated with enhanced feature amplitude, ensuring that the features are highly sensitive to arc events. Secondly, the arc fault multi-dimensional feature vector is designed as a five-dimensional arc fault vector, which comprehensively describes the essential characteristics of arc events from five dimensions: average enhanced amplitude, proportion of high amplitude events, spatiotemporal dispersion, polarity balance, and maximum transient event outbreak rate, while being insensitive to noise events. All features are normalized or relatively quantized to make their value range controllable, providing a unified quantitative benchmark for subsequent classification judgment.
[0077] Explanatory, in-window average enhancement amplitude characteristics This value is used to characterize the overall arc intensity of an event; the larger the value, the more significant the overall arc characteristic. It also indicates the proportion of high-amplitude events. This value characterizes the proportion of strong electric arc events; the higher the value, the higher the proportion of strong electric arc events. It also reflects the spatiotemporal dispersion characteristics. This is used to characterize the relative clustering characteristics of events in the spatiotemporal domain; electric arc events are continuously triggered within a local space, exhibiting strong spatial clustering and concentrated temporal distribution, corresponding to spatiotemporal dispersion characteristics. The values are relatively large; the noise events are spatially dispersed and temporally uniform, corresponding to spatiotemporal dispersion characteristics. The values are relatively small; polarity balance characteristics. This is used to characterize the proportion of the absolute difference between the number of positive and negative events within a sliding window to the total number of events; maximum transient event burst rate characteristic. It is used to characterize the transient burst characteristics of electric arc events.
[0078] Optionally, high amplitude threshold The value is set to 1.0, based on the following criteria: the high-amplitude threshold for pure noise events is typically less than 0.5, while the high-amplitude threshold for arc events is typically greater than or equal to 1.0; therefore, the high-amplitude threshold is set to... Setting it to 1.0 can effectively distinguish between arc events and noise events.
[0079] In this embodiment 1, a sliding window feature extraction technique is used to extract multidimensional features from an event set with enhanced feature amplitude. This transforms the discrete event stream with enhanced amplitude into a structured multidimensional quantized feature vector, realizing the transformation from raw data to fault identification features and providing core input for the final fault determination. Specifically, the arc fault multidimensional feature vector is designed as a five-dimensional arc fault vector. Through the mutual complementarity of different dimensions, it can comprehensively describe the average enhanced amplitude, the proportion of high-amplitude events, the spatiotemporal dispersion, polarity balance, and the maximum transient event outbreak rate of arc events. It is highly sensitive to early arcs and can effectively distinguish residual random interference noise, greatly improving the accuracy and robustness of subsequent identification. All features are calculated within a fixed sliding window, with minimal computational load and strong real-time performance, fully adapting to the real-time operation requirements of embedded hardware. The feature value range is controllable, providing stable input for subsequent adaptive determination.
[0080] Step 6: Based on the multi-dimensional feature vector of the arc fault, the early arc fault detection result of the device under test is obtained through multi-feature fuzzy fusion processing.
[0081] Interpretively, the multi-dimensional feature vector of arc fault output in step 5 is used for multi-feature fuzzy fusion to achieve accurate identification and severity classification of arc fault, and output early warning signals that meet the operation and maintenance needs of power plants, thus completing a complete closed loop from detection to early warning.
[0082] Specifically, the implementation process is as follows: Step 61: Calculate the membership degree of each feature in the multidimensional feature vector of arc faults based on the single-feature membership degree calculation function; the single-feature membership degree calculation function is as follows:
[0083] in, For the first Normalized membership degree of a feature; For the first The actual calculated value of the dimensional feature; For the first The minimum reference value for a dimensional feature; For the first The maximum reference value for a dimensional feature.
[0084] Explanatory, the first Normalized membership degree of dimensional features , used to characterize the The degree of support for arc faults by dimensional features; Minimum reference value of dimensional features and the Maximum reference value of dimensional features All of these are determined based on extensive laboratory arc test data and power plant field test data; optionally, the first Minimum reference value of dimensional features and the Maximum reference value of dimensional features The specific values are as follows: when At that time, the minimum reference value of the first-dimensional feature The maximum reference value of the first-dimensional feature ;when At that time, the minimum reference value of the second-dimensional feature The maximum reference value of the second-dimensional feature ;when At that time, the minimum reference value of the third-dimensional feature The maximum reference value of the third-dimensional feature ;when At that time, the minimum reference value of the fourth dimension feature The maximum reference value of the 4th dimension feature ;when At that time, the minimum reference value of the 5th dimension feature The maximum reference value of the 5th dimension feature .
[0085] Step 62: Based on the membership degree of each feature in the multidimensional feature vector of the arc fault, calculate the comprehensive membership degree of the arc fault using the arc fault combined membership degree calculation function; the comprehensive membership degree calculation function of the arc fault is as follows:
[0086]
[0087] in, The overall membership degree of arc faults; For the first Weight coefficients of dimensional features.
[0088] Explanatory, arc fault comprehensive membership The larger the value, the higher the probability of an arc fault and the more severe the fault; Weight coefficients of dimensional features This is used to adjust the contribution of different features in the comprehensive judgment; among them, the optimal weights are determined by the analytic hierarchy process based on a large amount of laboratory arc test data and power plant field test data; optionally, the weight coefficient of the first-dimensional feature... The weight coefficients of the second-dimensional feature The weight coefficients of the third-dimensional feature The weight coefficients of the fourth feature dimension The weight coefficients of the 5th dimension feature .
[0089] Step 63: Based on the comprehensive membership degree of the arc fault, and according to the preset arc fault determination logic, obtain the early arc fault detection results of the equipment under test; wherein, the preset arc fault determination logic is as follows:
[0090] in, Fault level; The first membership threshold; This is the second membership threshold; This is the third membership threshold.
[0091] Explanatory, fault level The value can be 0, 1, 2, or 3, corresponding to four different fault levels, determined by a piecewise function based on the comprehensive membership degree of the arc fault and the classification threshold. Each membership degree threshold is determined according to the power industry standard "Technical Supervision Regulations for Power Grid Metals" (DL / T 1424-2015) and in conjunction with the actual operation and maintenance needs of the power plant. Optionally, the first membership degree threshold... , representing the threshold separating the normal state from the attentional state; the second membership threshold. This represents the threshold separating the attentional state from the abnormal state; the third membership threshold. , which represents the threshold separating abnormal and emergency states.
[0092] It should be noted that after feature quantization is completed in step 5, accurate identification and graded early warning of arc faults are required. However, most existing technologies use single feature judgment with fixed thresholds, which are easily affected by the field environment, resulting in high false alarm rates and poor robustness. Secondly, existing technologies can only achieve binary judgment of the presence or absence of an arc, and cannot distinguish the severity of the fault, thus failing to meet the needs of power plant operation and maintenance. In addition, existing technologies cannot adaptively fuse complementary information of multi-dimensional features, resulting in insufficient ability to identify early arcs. In this embodiment 1, based on fuzzy comprehensive evaluation, the arc fault judgment logic is determined by combining the characteristic features of arc faults in power plants with multi-dimensional feature vectors of arc faults. Among them, the fuzzy comprehensive evaluation idea is adopted to determine the arc fault judgment logic. Multi-dimensional feature fusion is used to achieve comprehensive membership of arc faults, taking into full account the contribution of features from different dimensions. Compared with single threshold judgment, the accuracy and robustness are greatly improved. For the characteristics of each feature, a positive or negative membership function is designed to normalize features with different value ranges and characteristics to a membership interval of [0,1], realizing the effective fusion of multi-dimensional features. Feature weight coefficients optimized based on engineering experience are introduced, which can be adaptively adjusted according to different equipment types and field environments to adapt to different monitoring scenarios. Strictly following power industry standards, a four-level fault level judgment mechanism is designed, corresponding to different early warning levels and operation and maintenance actions, which is fully adapted to the operation and maintenance management process of power plants and has strong engineering practicality.
[0093] Optionally, the early arc fault detection results of the device under test in this embodiment 1 include the comprehensive membership degree of the arc fault. Fault Level And the corresponding graded early warning signals; wherein, the generation process of the corresponding graded early warning signals is shown in Table 1 below.
[0094] Table 1 shows the generation rules for the corresponding graded early warning signals.
[0095] In this embodiment 1, the accurate identification of arc faults is achieved through the fusion decision of multi-dimensional features, especially the high sensitivity identification of early arcs, which solves the problems of high false alarm rate and missed detection rate of existing technologies. Secondly, in accordance with the power industry standards, the severity of the fault is classified and the corresponding graded early warning signal is output, which directly adapts to the operation and maintenance management process of power plants and realizes a complete engineering closed loop from detection to early warning. In addition, by adjusting the weight coefficient and membership threshold, it can adapt to the monitoring needs of different power plant electrical equipment such as generators, transformers, and switchgear, and can also adapt to different field environments, with extremely strong generalization.
[0096] The early arc fault detection method for power plant electrical equipment described in Embodiment 1 achieves an early arc detection rate of greater than or equal to 95% in complex power plant environments, a false alarm rate of less than or equal to 5% in arc-free scenarios, and an alarm output delay of less than 50ms upon detection. It can also output tiered early warning signals, adapting to power plant operation and maintenance management processes. It boasts advantages such as strong resistance to industrial environmental interference, high sensitivity in early arc detection, low false alarm rate, strong real-time performance, and good engineering adaptability.
[0097] Example 2 As attached Figure 2 As shown in Embodiment 2, this invention provides an early arc fault detection system for power plant electrical equipment, used to implement the early arc fault detection method for power plant electrical equipment, including an event acquisition module, an event initial screening module, a noise event filtering module, a feature orientation enhancement module, a feature extraction module, and a fault determination module.
[0098] The event acquisition module is used to acquire the raw event set of the device under test; the raw event set is the raw event set output by the event camera, which contains several events arranged in chronological order.
[0099] The event initial screening module is used to remove ambient light unipolar pseudo-events from the original event set based on the bipolarity matching degree of the event's spatiotemporal neighborhood, thereby obtaining the initial screening effective event set.
[0100] The noise event filtering module is used to combine the spatiotemporal continuity of the electric arc to calculate the spatiotemporal joint credibility weight of each event in the initial screening of valid events, and to remove invalid noise events to obtain an event set with credibility weights.
[0101] The feature orientation enhancement module is used to perform early arc feature orientation enhancement on events in the event set with confidence weights based on the physical law of directional diffusion of electric arcs, and combined with the spatiotemporal joint confidence weight of each event in the event set with confidence weights, so as to obtain an event set with enhanced feature amplitude.
[0102] The feature extraction module is used to extract multidimensional features from an event set with enhanced feature amplitude using sliding window feature extraction technology, thereby obtaining a multidimensional feature vector of arc fault.
[0103] The fault determination module is used to output the early arc fault detection results of the equipment under test by processing the multi-dimensional feature vector of the arc fault through multi-feature fuzzy fusion.
[0104] Example 3 As attached Figure 3As shown, this embodiment 3 provides an electronic device, including: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the method for detecting early arc faults in power plant electrical equipment; or, the processor executing the computer program to implement the functions of each module in the above-mentioned system for detecting early arc faults in power plant electrical equipment.
[0105] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a preset function, the instruction segments describing the execution process of the computer program in the electronic device.
[0106] The electronic device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above are examples of electronic devices and do not constitute a limitation on the electronic device. It may include more components than described above, or combine certain components, or different components. For example, the electronic device may also include a communication interface, input / output devices, network access devices, and a bus.
[0107] The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor, or any conventional processor. The processor is the control center of the electronic device, connecting various parts of the electronic device via various communication interfaces and lines.
[0108] The memory can be used to store the computer program and / or module. The processor implements various functions of the electronic device by running or executing the computer program and / or module stored in the memory and by calling the data stored in the memory.
[0109] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function (such as sound playback, image playback, etc.). The data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart memory cards, secure digital cards, flash memory cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.
[0110] Example 4 This embodiment 4 also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for early arc fault detection of electrical equipment in a power plant.
[0111] If the modules / units integrated into the early arc fault detection system for the power plant's electrical equipment are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.
[0112] Based on this understanding, the present invention can implement all or part of the processes in the aforementioned method for detecting early arc faults in power plant electrical equipment, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the aforementioned method for detecting early arc faults in power plant electrical equipment. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or a preset intermediate form, etc.
[0113] The computer-readable storage medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal, and software distribution medium, etc.
[0114] Example 5 This embodiment 5 provides a computer product, which includes a computer program stored in a computer-readable storage medium. The processor of the electronic device reads the computer program from the computer-readable storage medium and executes the computer program, so that the electronic device can execute the early arc fault detection method for power plant electrical equipment described in embodiment 1, which will not be described again here.
[0115] It should be noted that those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above methods.
[0116] The early arc fault detection method for power plant electrical equipment described in this invention acquires the original event set output by an event camera. Based on the bipolar matching degree of the event's polarity spatiotemporal neighborhood, it eliminates unipolar pseudo-events caused by direct sunlight, equipment reflection, and strong electromagnetic interference in the power plant scene, completing the initial screening and noise reduction. Subsequently, combining the spatiotemporal continuity of the arc, it calculates the spatiotemporal joint credibility weight of each event in the effective event set of the initial screening to further eliminate invalid noise events, forming a two-level anti-interference noise reduction mechanism. This effectively reduces the false alarm rate and the false negative rate, ensuring that the false alarm rate in arc-free scenarios is less than or equal to 5% and the early arc detection rate is greater than or equal to 95%. Afterward, combining the physical laws of arc directional diffusion, it uses the spatiotemporal joint credibility weight of the event to perform directional feature enhancement on the arc event, effectively highlighting the microsecond-level early arc characteristics. To prevent the signal from being drowned out by background noise, true early fault warning is achieved. Subsequent use of sliding window feature extraction technology and multi-feature fuzzy fusion processing eliminates the need for complex deep learning models, ensuring a latency of less than 50ms from event acquisition to alarm output, meeting the real-time requirements of online monitoring. Furthermore, the entire detection process is designed based on the physical nature of arc discharge, rather than simple data fitting, exhibiting strong generalization capabilities and adaptability to various electrical equipment in power plants. The detection results can also meet the needs of power plant operation and maintenance management, providing reliable support for engineering implementation. This effectively solves the problems of poor generalization and inability to directly adapt to engineering applications in existing technologies, fundamentally improving the accuracy, anti-interference, real-time performance, and engineering practicality of early arc detection for power plant electrical equipment. It can effectively prevent equipment failures and major safety accidents caused by insulation degradation.
[0117] In this invention, a two-stage anti-interference mechanism is designed to address the specific interference in power plant scenarios. This mechanism consists of an initial screening based on the bipolar matching degree of the event-based polarity spatiotemporal neighborhood and a secondary denoising based on the event-based spatiotemporal joint credibility weight. This mechanism can effectively eliminate direct sunlight, equipment reflection, and strong electromagnetic interference in power plant scenarios, exhibiting extremely strong resistance to industrial environmental interference and a low false alarm rate. Secondly, based on the physical nature of arc discharge, early arc feature directional enhancement is performed, which can effectively identify early arcs at the microsecond level, achieving true early fault warning. This mechanism has extremely high sensitivity for early arc detection and a low false alarm rate.
[0118] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.
Claims
1. A method for detecting early arc faults in electrical equipment of a power plant, characterized in that, include: Obtain the raw event set of the device under test; wherein, the raw event set is the set of raw events output by the event camera, which contains several events arranged in chronological order; Based on the bipolar matching degree of the polarity spatiotemporal neighborhood of the event, ambient light unipolar pseudo-events in the original event set are removed to obtain a preliminary set of valid events; By combining the spatiotemporal continuity of the electric arc, the spatiotemporal joint credibility weight of each event in the initial screening effective event set is calculated, and invalid noise events are removed to obtain an event set with credibility weights. Based on the physical laws of the directional diffusion of electric arcs, and combined with the spatiotemporal joint credibility weight of each event in the event set with credibility weight, early electric arc feature directional enhancement is performed on the events in the event set with credibility weight to obtain an event set with enhanced feature amplitude. By using the sliding window feature extraction technique, multidimensional feature extraction is performed on the event set with enhanced feature amplitude to obtain the multidimensional feature vector of arc fault; Based on the multidimensional feature vector of arc faults, the early arc fault detection results of the device under test are obtained through multi-feature fuzzy fusion processing. Based on the bipolar matching degree of the event's spatiotemporal neighborhood, the process of removing ambient light unipolar pseudo-events from the original event set to obtain the initial screening of valid event sets is as follows: Calculate the bipolarity matching degree of the spatiotemporal neighborhood of each event in the original event set; The bipolarity matching degree of the spatiotemporal neighborhood of each event is compared with the preset matching degree threshold. The ambient light unipolar pseudo-events in the original event set are removed to obtain the initial set of effective events. The process of calculating the bipolar matching degree of the polarity spatiotemporal neighborhood for each event in the original event set is as follows: in, For the first The bipolar matching degree of the spatiotemporal neighborhood of an event; For the first Spatial coordinates of an event Centered on, spatial neighborhood radius Within, and the timestamp is within the time window. Inside, with the The total number of events with opposite polarities; For the first The trigger timestamp of each event; The time neighborhood is half the width; For the first Spatial coordinates of an event Centered on, spatial neighborhood radius Within, and the timestamp is within the time window. Within, the total number of events of all polarities; To prevent the removal of the zero smoothing factor; The process of calculating the spatiotemporal joint confidence weight of each event in the initial screening of valid events, taking into account the spatiotemporal continuity of the electric arc, and removing invalid noise events to obtain the event set with confidence weights, is as follows: Based on the bipolar matching degree of the spatiotemporal neighborhood of each event in the initial screening effective event set, the spatiotemporal joint credibility weight of each event in the initial screening effective event set is calculated using the Gaussian credibility weighted model constrained by the spatiotemporal neighborhood. The spatiotemporal joint credibility weight of each event is compared with the preset credibility weight threshold, and invalid noise events are eliminated to obtain a set of events with credibility weights. The Gaussian credibility-weighted model with spatiotemporal neighborhood constraints is as follows: in, For the first Spatiotemporal joint credibility weights of events; For the first The spatiotemporal neighborhood set of an event; For the first The spatiotemporal neighborhood set of an event The first The sequence number of each event; For the first The bipolar matching degree of the spatiotemporal neighborhood of an event; For the first The spatiotemporal neighborhood set of an event Total number of events within; To prevent the removal of the zero smoothing factor; The standard deviation of the Gaussian kernel in space; The standard deviation of the Gaussian kernel over time; For the first The ordinate of the pixel coordinate system for each event; For the first The ordinate of the pixel coordinate system for each event; For the first The trigger timestamp of each event; For the first The trigger timestamp of each event; For the first The x-coordinate of the pixel coordinate system for each event; For the first The ordinate of the pixel coordinate system for each event; The process of early arc feature-oriented enhancement for events in a set of events with confidence weights is as follows: in, For the first The amplitude of the enhanced features of each event; For the first Spatiotemporal joint credibility weights of events; The feature enhancement coefficient; For the first The set of spatiotemporal evolution neighborhoods of an event; For the first The spatiotemporal evolution neighborhood of an event The first The sequence number of each event; For the first Spatiotemporal joint credibility weights of events; For from the first The event to the The unit motion direction vector of each event; is the typical unit direction vector of arc diffusion in the device under test; To prevent the elimination of the zero smoothing factor.
2. The method for detecting early arc faults in power plant electrical equipment according to claim 1, characterized in that, The process of using sliding window feature extraction technology to extract multidimensional features from an event set with enhanced feature amplitudes to obtain a multidimensional feature vector of arc faults is as follows: in, This represents the multidimensional feature vector of an arc fault. This represents the average enhancement amplitude characteristic; The characteristic is the proportion of high-amplitude events; It is characterized by spatiotemporal dispersion; It is characterized by polarity balance; This is a characteristic of the maximum transient event burst rate; For event sets with enhanced feature amplitude Total number of events; For a set of events with enhanced feature amplitude; For the first The amplitude of the enhanced features of each event; For the sliding window to satisfy The number of events; To prevent the removal of the zero smoothing factor; The standard deviation of all event space coordinates within the sliding window; The standard deviation of all event timestamps within the sliding window; Polarity within the sliding window The number of events; Polarity within the sliding window The number of events; For time Within the time sub-window of the center's statistics, the following conditions are met. The number of events; High amplitude threshold; The width of the time sub-window is used for statistics.
3. The method for detecting early arc faults in power plant electrical equipment according to claim 1, characterized in that, The process of obtaining the early arc fault detection results of the device under test by using the multi-dimensional feature vector of arc faults and multi-feature fuzzy fusion processing is as follows: Calculate the membership degree of each feature in the multidimensional feature vector of arc faults; The comprehensive membership degree of arc fault is calculated based on the membership degree of each feature in the multidimensional feature vector of arc fault. Based on the comprehensive membership degree of arc faults, and according to the preset arc fault judgment logic, the early arc fault detection results of the equipment under test are obtained.
4. A system for detecting early arc faults in power plant electrical equipment, characterized in that, A method for detecting early arc faults in power plant electrical equipment as described in any one of claims 1-3, comprising: The event acquisition module is used to acquire the raw event set of the device under test; wherein, the raw event set is the raw event set output by the event camera, which contains several events arranged in chronological order; The event screening module is used to remove ambient light unipolar pseudo-events from the original event set based on the bipolarity matching degree of the event's spatiotemporal neighborhood, thereby obtaining a preliminary set of valid events. The noise event filtering module is used to combine the spatiotemporal continuity of the electric arc to calculate the spatiotemporal joint credibility weight of each event in the initial screening of valid events, and to remove invalid noise events to obtain an event set with credibility weights. The feature orientation enhancement module is used to perform early arc feature orientation enhancement on events in the event set with confidence weights based on the physical law of directional diffusion of electric arcs, and combined with the spatiotemporal joint confidence weight of each event in the event set with confidence weights, to obtain an event set with enhanced feature amplitude. The feature extraction module is used to extract multidimensional features from an event set with enhanced feature amplitude using sliding window feature extraction technology, thereby obtaining a multidimensional feature vector of arc fault. The fault determination module is used to output the early arc fault detection results of the equipment under test by processing the multi-dimensional feature vector of the arc fault through multi-feature fuzzy fusion.
5. An electronic device, characterized in that, include: A processor is used to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, performs the method for early arc fault detection of power plant electrical equipment as described in any one of claims 1-3.
6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for early arc fault detection of electrical equipment in power plants as described in any one of claims 1-3.
7. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method for early arc fault detection of power plant electrical equipment as described in any one of claims 1-3.
Citation Information
Patent Citations
Event stream denoising method based on adaptive time-space weight
CN121051339A
Method and system for carrying out real-time lightning classification by utilizing dynamic visual data
CN121169972A
