Power distribution network line fault detection method and system based on data analysis

Abstract

The invention relates to the technical field of power distribution network detection, in particular to a power distribution network line fault detection method and system based on data analysis. Establishing a superposition model and a time-varying capacitance model according to a water film capacitance effect and wind vibration interference, determining a discharge peak amplitude and an exponential attenuation constant, measuring the distortion degree of water film coupling to a pure discharge signal and the interference degree of ground coupling to the pure discharge signal according to a water film trailing coefficient and a wind vibration suppression coefficient, and determining a time domain feature; using a sparse reconstruction algorithm to separate and calculate a discharge source number; thermal stress inhibition acoustic energy is obtained by combining thermal stress crack noise and discharge sound confusion caused by environment temperature change; dynamically adjusting a pulse sliding window according to the water film thickness change influence, and determining a pulse occurrence rate and a phase window ratio; and establishing a feature vector through the time domain feature, the discharge source number, the thermal stress suppression acoustic energy, the pulse occurrence rate and the phase window ratio to realize the detection of the cable arc fault or the surface partial discharge.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network detection, and particularly to a method and system for detecting faults in distribution network lines based on data analysis. Background Art

[0002] In the actual operation environment of the distribution network, two complex and superimposed climate and environmental factors are often encountered: on the one hand, in strong wind weather conditions (such as the frequent windy weather in Gansu region), the wire will bend and swing, generating mechanical vibration and sudden change of the conductor-ground capacitance; on the other hand, in the same area, especially during occasional seasonal rainfall, frost, or rapid change of the relative temperature at night, a thin water film will form on the surface of the insulation components of the distribution line (such as suspension porcelain insulators, constantan joints, etc.) on the dirt or salt spray deposition layer; these two environmental factors will respectively produce different types of interference on the discharge signals measured by the high-frequency current transformer (HFCT), such as: 1) Most of the current distribution network line fault detection systems use HFCT sensors or UHF antennas to monitor sudden line discharges - whether it is a line arc fault Arc or a surface partial discharge PD, both will cause high-frequency transient pulses. Traditional methods often only focus on the time-domain rise / decay characteristics or the frequency-domain energy distribution of the signals, but do not fully consider the comprehensive influence of the dirt or salt spray deposition layer. In fact, when a water film is generated on the pollution layer, when a partial discharge or a line arc occurs on the surface of the insulator or the constantan joint, part of the discharge energy will be absorbed by the water film and continuously change the capacitance value during its evaporation process; 2) The salt content and pollutant ion conductivity in the water film are very strong, and it can conduct electricity briefly near the voltage peak value, and then quickly evaporate to restore the capacitance. In this way, in addition to the pure exponential decay of the current waveform monitored by the HFCT, a slow tail wave of the capacitance coupling trailing type will be superimposed, usually manifested as a signal double peak or trailing; 3) The time-domain distortion of the water film trailing will make the traditional algorithms that rely on the peak amplitude and the exponential decay constant to distinguish weak Arc and PD inaccurate. For example, when the trailing component of the water film exceeds a certain amplitude, the arc waveform that has already decayed rapidly may be misjudged as multiple high-speed PDs, or an abnormally large decay constant value may be fitted from the decay curve of the PD signal, thus causing false alarms or missed alarms; Simultaneously with the water film interference, when the wind speed exceeds the threshold value, the elasticity and tension of the wire will cause its high-frequency swing and vibration. On the one hand, this large wind swing will generate mechanical noise in the insulator (ultrasound, and the acoustic channel will pick up the corresponding clutter), and on the other hand, more critically - the instantaneous offset of the wire relative to the ground or nearby metal objects will cause a sudden change in the conductor-ground capacitance; Therefore, it is currently difficult to combine the two complex interferences of the water film capacitance trailing and the conductor-ground capacitance coupling interference in the strong wind environment to realize the discrimination and detection of faults in the distribution network. Summary of the Invention

[0003] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method and system for detecting faults in distribution network lines based on data analysis, which can effectively solve the problem that the existing technology fails to consider the complex interference effects of the water film capacitance tailing in the air environment and the guide-ground capacitance coupling interference in the strong wind environment on the fault discrimination of the distribution network, where these two types of interference effects are superimposed on each other.

[0004] To achieve the above objectives, the present invention is realized through the following technical solutions: The present invention provides a method and system for detecting faults in distribution network lines based on data analysis, including at least: Establish a superposition model and a time-varying capacitance model based on the water film capacitance effect and wind vibration interference, determine the discharge peak amplitude and the exponential decay constant, and respectively measure the distortion degree of the water film coupling on the pure discharge signal and the interference degree of the guide-ground coupling on the pure discharge signal according to the water film tailing coefficient and the wind vibration suppression coefficient, so as to establish the time-domain characteristics; Introduce the non-uniform state of the pollution layer and use the sparse reconstruction algorithm to separate and calculate the number of discharge sources; Combined with the confusion between the thermal stress crack noise caused by the change of the environmental temperature and the discharge sound, use the environmental temperature to suppress the thermal stress interference and obtain the thermal stress suppressed acoustic energy; Dynamically adjust the pulse sliding window according to the change of the water film thickness and determine the pulse incidence rate and the phase window ratio; Establish a feature vector through the above-mentioned time-domain characteristics, the number of discharge sources, the thermal stress suppressed acoustic energy, the pulse incidence rate and the phase window ratio, and realize the detection and judgment of the arc fault or surface partial discharge of the distribution network cable according to the logistic regression model.

[0005] By simultaneously fitting the exponential decay discharge, the water film capacitance tailing and the wind vibration coupling spike in a single pulse time-domain window, the characteristics of the pure discharge signal can be accurately extracted in a complex environment, and misjudgment will not occur due to the failure of a single interference model; In an environment with both high salt fog and strong wind, if it is raining and the water film thickness is the largest just after the rain, local discharge is more likely to occur at this time. The traditional method often ignores the superposition effect. However, the present invention determines that the water film tailing is dominant through the water film tailing coefficient, then eliminates this component, and then eliminates the wind vibration coupling judged by the wind vibration suppression coefficient from the remaining waveform. Finally, the remaining waveform is the closest to the pure arc discharge or pure PD waveform, making the subsequent multi-source OMP positioning and feature fusion discrimination more accurate; In terms of the discrimination logic, by combining the water film tailing coefficient and the wind vibration suppression coefficient, it is possible to avoid wasting the computational effort for fitting the wind vibration term in the subsequent process, save the online computational effort and improve the detection efficiency; The introduction of the water film trailing coefficient and the wind vibration suppression coefficient allows for directly skipping the complete joint fitting when the interference of most pulse signals is relatively light, and only extracting the original time-domain exponential decay term. Only when it is determined that the interference is significant, the wind vibration or water film is removed before performing the non-linear least squares fitting. This avoids the situation where all pulses require time-consuming double-exponential or triple-term superposition fitting for in-depth optimization, and can actually shorten the average fitting time for each pulse in practice.

[0006] The method for establishing the superposition model is as follows: Define the triggering threshold of the high-frequency current transformer HFCT , if there exists: When the amplitude of the current signal and the interval from the previous trigger is greater than or equal to the minimum sampling point interval, record a pulse, and the trigger sampling sequence number is ; Extract the pulse time-domain window: 、 respectively represent the number of sampling points on both sides of the window width, represents the intercepted signal within the time-domain window where the th trigger event is located, represents the time-domain sequence of the pulse current of the original HFCT sampling; Establish the superposition model: represents the peak amplitude of the pure discharge of the th pulse, represents the exponential decay constant, represents the transient reference voltage on the insulator surface, represents the time corresponding to the th sampling point within the pulse window, represents the instantaneous equivalent capacitance corresponding to the time of the th sampling point, represents the time corresponding to the th sampling point of the instantaneous capacitance between the lower conductor and the ground, represents the line voltage value at the th sampling point.

[0007] Furthermore, based on the water film capacitance effect and wind vibration interference, judge the surface partial discharge or line arc fault according to the water film trailing coefficient and the wind vibration suppression coefficient .

[0008] Furthermore, the method for obtaining the thermal stress to suppress the acoustic energy is as follows: Temperature gradient Calculation: represents the time interval of ambient quantity sampling, represents the ambient temperature corresponding to the th pulse trigger time; Calculate the ultrasonic energy during the th pulse : represents the acoustic signal value at the th sampling point; Define the thermal stress acoustic suppression coefficient represents the reference temperature, represents the temperature sensitivity coefficient; If , are respectively greater than the corresponding thresholds, it is determined that the ultrasonic energy comes from thermal stress crack noise, and the thermal stress suppressed acoustic energy is obtained.

[0009] Furthermore, the method for determining the pulse incidence rate is: Count the number of triggers in the time period represents the index of the trigger event, represents the discrete sampling point number corresponding to the th pulse trigger, represents the sampling rate of the high-frequency current transformer, represents the dynamic sliding window length corresponding to the th trigger; represents the counting operation of the number of set elements; Calculate the pulse incidence rate according to the ratio of the counted number of triggers to the dynamic sliding window length.

[0010] Furthermore, the method for the logic regression model to detect cable arc faults or surface partial discharges is: Construct a logic regression model and input the feature vector, and output the arc discharge probability: Construct a cross-entropy loss function; Calculate the arc discharge probability based on the feature vector obtained each time; If it is greater than the probability threshold, it is determined as a cable arc fault, otherwise it is a surface partial discharge.

[0011] The technical solution provided by the present invention has the following beneficial effects compared with the known prior art: By simultaneously fitting the exponential decay discharge, the water film capacitance tail and the wind vibration coupling spike within a single pulse time domain window, the characteristics of the pure discharge signal can be accurately extracted in a complex environment, and misjudgment will not occur due to the failure of a single interference model; by judging that the water film tail is dominant through the water film tail coefficient, and then removing this component, and then removing the wind vibration coupling judged by the wind vibration suppression coefficient from the remaining waveform, the finally retained is the waveform closest to the pure arc discharge or pure PD waveform, making the subsequent multi-source OMP positioning and feature fusion discrimination more accurate.

[0012] According to the water film capacitance tail and the wind vibration coupling model, the clean pure discharge peak and decay characteristics are extracted, and the multi-source discharge positioning is carried out on the sparse array by using the HFCT signal after removing interference. The filtering bandwidth is adaptively adjusted in combination with the EIS aging index to ensure the capture of long-tail low-frequency and high-frequency signals. The thermal stress crack noise is suppressed based on the temperature gradient, the ultrasonic characteristics are optimized, and the statistical window is dynamically adjusted to ensure the reliability of the pulse frequency and phase statistics. Through the synergistic effect of each step and feature fusion, the arc fault and surface discharge can be accurately distinguished under humid, windy and heavily polluted conditions, improving the detection accuracy and real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0014] Figure 1 It is a schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] In order to make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0016] In the traditional detection of distribution network line faults, especially the high-frequency discharge detection based on HFCT sensors, the main concern is the distinction between cable arc discharge (Arc) and surface partial discharge (PD) in transient pulse signals. Early research and engineering practices are often based on the following situations: When there is no obvious water film on the insulator surface and the contamination layer is dry, the discharge signal mainly appears as an exponentially decaying high-frequency pulse. In this case, traditional methods generally only need to perform simple time-domain exponential fitting to accurately extract the peak amplitude and exponential decay constant. This can then be combined with phase statistics, vibration signals, or fiber temperature monitoring to distinguish between Arc and PD. In windy conditions, conductor swaying causes interference from mechanical vibration and electromagnetic coupling signals. Many solutions use additional wind speed sensors or vibration sensors to decouple vibration noise. However, most methods only address acoustic interference or mechanical vibration, and fail to fully consider how changes in conductor-ground capacitance can directly introduce high-frequency spikes into the HFCT signal. Consequently, wind-induced electromagnetic coupling is often overlooked or misinterpreted as weak arc discharges. Distribution lines in some areas are subject to high pollution year-round. When pollutants (such as dirt, chemical fumes, and salt spray) deposit on the surfaces of conductor insulators or constantan connectors, they absorb atmospheric water vapor, forming a water film, particularly in humid weather. This water film generates surface partial discharge (PD) at the moment of discharge, accompanied by a sudden change in capacitance. This can manifest as tailing or "double-peak" distortion in the HFCT signal. Some studies have attempted to describe this using surface discharge models, but most methods rely solely on empirical threshold screening at the back end, making it difficult to directly incorporate transient changes in the water film coupling capacitance into the model. In real-world scenarios, wind vibration and water film often occur simultaneously. For example, in industrial areas, dense fog and strong winds can cause both water film capacitance tailing and high-frequency spikes from wind vibration in the HFCT signal. If these signals are processed separately, even after removing the water film tailing, misjudgment due to the wind vibration spikes may still occur. Alternatively, even after removing the wind vibration, the residual water film tailing may affect the exponential fitting. In addition, distribution network fault detection has strict real-time requirements. Once an Arc occurs on the line, which is a typical fault signal, an alarm must be issued and the protection strategy must be activated within milliseconds or less. Frequent large-scale data fitting to remove interference is not only time-consuming but also prone to delays or misjudgments.

[0017] In summary, current technical solutions are unable to simultaneously and quantitatively address the complex interference caused by water film capacitance tailing in air environments and ground capacitance coupling interference in windy environments. They can often only perform local optimization for a single environment, resulting in the following problems: In windy and dry environments, water film trailing is ignored and PD is missed; In the wet and windless environment after seasonal rain or frost, ignoring wind-vibration coupling can lead to misjudgment; When both exist at the same time, a single model often cannot make the HFCT fitting converge to a reasonable solution, and may even lead to misjudgment of the secondary model, and small Arcs are treated as common PDs, thus delaying fault isolation.

[0018] Based on the above analysis, the present invention is proposed, and the present invention will be further described below in conjunction with embodiments.

[0019] Embodiment 1 (refer to Figure 1 ): A method for detecting faults in distribution network lines based on data analysis, including: According to the environmental state of the distribution line, judge the current operating state of the distribution network, then there is: If in rainy days, after rain, etc., that is, in the existing air environment, the capacitance coupling effect caused by the instantaneous conduction and evaporation of the water film on the pollution layer (the dirt or salt spray, etc. attached to the insulators of the distribution line, such as the surface of suspension insulators or constantan joints) will distort the discharge pulse waveform (appearing as double peaks or trailing distortion). This kind of distortion seriously affects the extraction of signal features, especially the accuracy of the peak amplitude and exponential decay constant of the discharge signal, and misleads the judgment of line arc discharge and surface partial discharge (PD); Secondly, when there is strong wind, the instantaneous change of the conductor-ground capacitance between the conductor and the ground will cause high-frequency sharp pulse signals, which are manifested as broadband interference. This is similar to weak line arc discharge (Arc) or multi-source surface partial discharge PD signals, resulting in misjudgment. Especially when the conductor swings, it will generate a large transient coupling current, disturbing the HFCT signal, and it is difficult to distinguish the pulses caused by arc discharge and wind vibration. These interference signals may cause misjudgment, such as misjudging the wind vibration signal as a line arc discharge, or misjudging PD due to the influence of the water film capacitance. Therefore, design the calculation of the trailing coefficient of the water film capacitance and the wind vibration suppression coefficient to quantify the interference effect, and then more accurately extract the pure arc discharge signal. The specific steps are as follows: Define the trigger threshold of the high-frequency current transformer (HFCT, installed at the bottom of the insulator or on the wire busbar) , in the continuous HFCT data stream, if there is: When the current signal amplitude at the th sampling point and the previous trigger interval (the minimum sampling point interval), then it is considered that a discharge pulse event occurs at this moment, record a pulse once, and record the trigger sampling sequence number as ; Extract the pulse time domain window: , respectively represent the number of sampling points on both sides of the window width, represents the th trigger event (that is, the th discharge pulse) the intercepted signal within the time domain window, represents the Sampling point sequence number at the time of the sub-pulse trigger, Indicates the time-domain sequence of pulsed current obtained by sampling the original HFCT (High Frequency Current Transformer) (including the true discharge current (arc current), transient current caused by air breakdown or insulation damage; displacement current and capacitive current due to the time-varying capacitance effect formed by the water film on the pollution layer; coupling current formed by the change of the conductor-ground capacitance caused by the wind vibration of the conductor, which is also a kind of capacitive current); Establish a superposition model: Represents the pure discharge signal term, Represents the peak amplitude of the pure discharge of the sub-pulse, Represents the exponential decay constant, corresponding to the decay of the discharge energy, Represents the trailing term of the water film capacitance, Represents the transient reference voltage on the insulator surface, Represents the time corresponding to the th sampling point within the pulse window; Represents the instantaneous equivalent capacitance corresponding to the time of the th sampling point, Represents the wind vibration coupling term; , Represents the time corresponding to the th sampling point instantaneous capacitance between the lower conductor and the ground,[[ID=3*]] Represents the static conductor-ground capacitance after recovery in the absence of wind or wind vibration balance, Represents the relative extra conductor-ground capacitance added when the relative position or attitude of the conductor changes instantaneously due to wind vibration, Represents the exponential rate constant for the conductor-ground capacitance to return to the static value, , Represents the time corresponding to the th sampling point under the line voltage value, Represents the power grid fundamental frequency, Represents the line peak voltage; Establish a time-varying capacitance model: , Represents the water film capacitance at the moment of, Represents the reference capacitance when the pollution layer is not saturated, Represents the additional capacitance after the formation of the water film, Represents the water film evaporation rate constant, represents a real number variable (unit: seconds), describing the moment when the capacitance is checked. Indicates the The trigger moment is used here to describe the The triggering time of the secondary discharge pulse helps fit the HFCT signal, especially the tail component that appears in the signal, by simulating the capacitance change caused by the water film during the arc discharge process. By modeling the water film capacitance and voltage change rate, this model can accurately extract the time domain characteristics of the discharge signal, such as peak amplitude and exponential decay constant, thereby improving the accuracy of the arc discharge signal.

[0020] Estimation by least squares fitting or nonlinear least squares method 、 、 as well as (I will not elaborate on this here); Calculate the water film tailing coefficient , which measures the degree of distortion of the pure discharge signal caused by water film coupling:

[0021] Therefore, if If it is greater than the first preset threshold, it means that the water film tailing effect is serious and tends to be surface partial discharge (PD); Calculate wind vibration suppression coefficient , which measures the degree of interference of the conductive-ground coupling on the pure discharge signal:

[0022] like If the value is greater than the second preset threshold, it is determined that the wind vibration interference is significant, and the corresponding interference should be eliminated before extracting the time domain features. Therefore: like is greater than a first preset threshold, If the value is less than or equal to the second preset threshold, it is judged that the water film tailing is serious, the wind vibration interference can be ignored, and the pulse is judged to be a surface PD. The time domain characteristics are output. 、 、 、 ; like is less than or equal to a first preset threshold, If the value is greater than the second preset threshold, it is judged that the wind vibration interference is serious and the water film tail is light. The wind vibration term is first removed and then fitted. The corrected time domain characteristics are obtained by fitting according to the least squares method. 、 、 、 , I will not go into details here.

[0023] In summary, by introducing the water film capacitance effect and wind vibration interference correction, the distortion problem of HFCT signals caused by the water film effect and wind vibration electromagnetic interference in strong wind environments is solved, providing a more accurate method for distinguishing line arc discharge and surface partial discharge (PD) signals. By jointly modeling interference terms and quantifying the interference impact, this solution can improve the accuracy of time domain feature extraction, avoid misjudgment in traditional methods, and enhance the overall performance of the system, especially its adaptability in complex environments.

[0024] 2) When the contamination layer is non-uniform, multiple discharge sources (surface partial discharge (PD) or line arc fault (Arc)) may be located on the same insulator. Single-channel or dual-channel positioning may experience aliasing (multiple surface partial discharge (PD) sources will generate superimposed signals, resulting in multiple solutions or deviations in one-way positioning). A sparse reconstruction algorithm is used to separate and locate multiple discharge sources and distinguish between multiple PDs and single-point Arcs, including: Array observation matrix construction: Set the total number of array channels , mixed observation vector ; Discrete the suspected discharge position of the insulator surface and the conductor into candidate points ; Constructing the observation matrix , No. Column indicates if at position When a discharge occurs, the delay and attenuation characteristics of the discharge signal in each channel; OMP sparse reconstruction solution: For the first The observation value vector within the sub-pulse window , represents transpose, Indicates the The sensor channels are in sequence number The sample value at Assume there is a sparse source vector ,satisfy: , represents the residual tolerance, Indicates the The residual vector of the trigger event, represents the observation matrix; Solved using the orthogonal matching pursuit algorithm , Indicates that the constraints listed below are met and a sparse coefficient vector is obtained , calculate the discharge source ; Among them, if Greater than 1 and all ( If the tops of the index sets (indicating the index sets) all fall in the high region, then this event tends to be multi-source surface PD; If it is equal to 1 and the corresponding candidate point position is close to the middle of the conductor ( An element in the candidate point set, representing the th candidate discharge position, describing the theoretically possible discharge positions, indicating the actual candidate point position corresponding to the non-zero component obtained by sparse reconstruction. It points to a specific candidate point position, corresponding to the most likely discharge source position in the non-zero sparse coefficient), that is, not on the insulator surface, then it tends to be single-point Arc.

[0025] 3) Dramatic changes in ambient temperature can cause thermal stress crack noise, generating ultrasonic pulses similar to PD / Arc. Specifically, the temperature difference between day and night or the thermal expansion difference between the metal constantan joint and the ceramic insulator will produce micro-crack friction or thermal stress acoustic pulses. The spectrum covers the ultrasonic and low-frequency acoustic regions, confusing with the PD acoustic signal. The ultrasonic channel cannot distinguish the crack friction sound from the PD sound, resulting in distorted acoustic features. Therefore, these interferences are suppressed through temperature information. The specific steps are as follows: Execute the temperature gradient Calculate: , if is greater than the thermal stress threshold , then there may be crack sound at this moment, represents the time interval for sampling environmental quantities (temperature, humidity, etc.), represents the th ambient temperature corresponding to the pulse trigger moment, represents at the th pulse trigger moment the ambient temperature before; Calculate the ultrasonic energy during the th pulse (i.e., the energy integral of the acoustic signal collected in the ultrasonic sensing channel): , represents the acoustic signal value collected by the ultrasonic sensor at the th sampling point; Define the thermal stress acoustic suppression coefficient , represents the reference temperature, represents the temperature sensitivity coefficient; If it satisfies , represents the suppression threshold, then it is considered that the ultrasonic energy mainly comes from thermal stress crack noise, and it is suppressed to obtain the thermal stress suppressed acoustic energy : This correction ensures that the acoustic energy in subsequent eigenvectors is not interfered with by thermal stress noise. Therefore, the comparison between the comprehensive temperature gradient and the ultrasonic energy itself can effectively distinguish between thermal stress crack friction sound and discharge sound when the temperature changes sharply. Secondly, when there is a large temperature difference between day and night or when there is a difference in thermal expansion between the constantan connector and the ceramic insulator, the outbreak of thermal crack sound can be predicted in advance, allowing the system to actively suppress the ultrasonic characteristics during this period to prevent them from being mistakenly called for PD / Arc judgment.

[0026] 4) Furthermore, after rainfall or dew, the rapid change in water film thickness can cause the pulse triggering timing to drift, destroying the original pulse frequency and phase concentration indicators. Pulse density and phase statistics based on a fixed window width are prone to pseudo-random distribution and misjudgment as PD. Therefore, the statistical window is dynamically adjusted according to the change rate of the water film thickness to ensure the reliability of the statistical sample. The specific steps are as follows: Collect water film thickness change rate (When calculating its value, the sampling time interval is ), calculate the dynamic sliding window length , Indicates the initial sliding window, such as 1s, Represents the water film change sensitivity coefficient, if If the value is large, increase the window length to ensure that there are still enough trigger events to count after nanosecond drift. Pulse rate statistics: Count the number of triggers in a time period , Indicates the index of the triggering event, Indicates the The discrete sampling point number corresponding to the pulse trigger, Indicates the high-frequency current transformer sampling rate, Indicates the The length of the dynamic sliding window corresponding to the trigger, The counting operation representing the number of elements in a set is equivalent to the cardinality of the set; Calculate the pulse rate ; Use grid phase calculation Phase window ratio at the second trigger , grid fundamental frequency , Indicates the The second trigger moment, describing the phase window ratio in the The moment of the second trigger (in this solution, Both represent trigger moments and are only used to describe the time point of each event or measurement in the system). Indicates the reference initial phase angle of the grid phase, and then the water film change rate Automatically widen the window when it is severe ,ensure a sufficient number of trigger samples so that pulse rate statistics and phase concentration are performed on a more balanced time base, avoiding pseudo-random distribution caused by microsecond drift; Secondly, combining the pulse occurrence rate and phase window ratio of the adaptive window can more reliably distinguish between continuous multiple PD triggers and occasional Arc triggers.

[0027] 5) Therefore, the time domain features calculated above ( 、 、 、 )、Discharge source number( ), thermal stress suppression of acoustic energy ( ), pulse rate ( ) and the phase window ratio ( ) Fusion to create feature vector , by building a logistic regression model, the arc discharge probability is output : , represents the weight vector of the logistic regression model, Represents the bias term of the logistic regression model; Constructing the cross entropy loss function , , Indicates the sample label (Arc is 1, PD is 0), which is solved by gradient descent or Adam optimizer , I will not elaborate on this here, and then, based on each obtained calculate If it is greater than the probability threshold, it is determined to be a cable arc fault Arc, otherwise it is a surface partial discharge PD. Therefore, the judgment is performed: If a line arc fault occurs, a fault alarm is generated and reported to the control center, triggering the isolation and positioning module (based on travel wave positioning or cross-correlation positioning) to quickly remove the faulty section, triggering the network self-healing (reconstruction) strategy, and restoring power supply to unaffected users; If partial discharge occurs on the surface, the number of PD events of the insulator is accumulated. If the number of PD events exceeds the threshold within a certain period of time, a PD alarm is generated, prompting the operation and maintenance to perform on-site cleaning or take anti-pollution measures.

[0028] In summary, based on the water film capacitance trailing term and the wind vibration coupling model, the present invention extracts clean pure discharge peaks and attenuation characteristics, uses the HFCT signal after removing interference to perform multi-source discharge positioning on a sparse array, adaptively adjusts the filtering bandwidth in combination with the EIS aging index to ensure the capture of long-tail low-frequency and high-frequency signals, suppresses thermal stress crack noise based on the temperature gradient, optimizes ultrasonic characteristics, dynamically adjusts the statistical window to ensure reliable pulse frequency and phase statistics, and through the synergistic effect of each step and feature fusion, accurately distinguishes arc faults and surface discharges under strong wind and heavy pollution conditions based on feature fusion, improving the detection accuracy and real-time performance.

[0029] Finally, the present invention also provides: A distribution network line fault detection system implemented according to the distribution network line fault detection method based on data analysis described above; A computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method are implemented; A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method are implemented; details are not repeated here, and reference may be made to the above method.

[0030] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting faults in distribution network lines based on data analysis, characterized in that, It includes the following steps: Monitor the amplitude of the current signal, and determine whether to extract the pulse time-domain window to obtain the intercepted signal within the time-domain window where the trigger event is located; In response to the input intercepted signal, construct a pure discharge signal term based on the pure discharge peak amplitude and the exponential decay constant, construct a water film capacitance trailing term based on the instantaneous equivalent capacitance, the reference voltage, and the capacitance change rate, and construct a wind vibration coupling term based on the instantaneous capacitance between the wire and the ground and the grid voltage. Thus, establish a superposition model and solve for the pure discharge peak amplitude and the exponential decay constant; Calculate the maximum deviation of the water film capacitance trailing term and the maximum amplitude of the wind vibration coupling term, and respectively combine them with the pure discharge peak amplitude to sequentially determine the water film trailing coefficient and the wind vibration suppression coefficient; Construct time-domain features; If the pollution layer is in a non-uniform state, use a sparse reconstruction algorithm to separate and calculate the number of discharge sources; Combined with the thermal stress crack noise and discharge sound confusion caused by the change in ambient temperature, use the ambient temperature to suppress the thermal stress interference to obtain the thermal stress suppressed acoustic energy; Collect the initial sliding window, dynamically adjust the length of the pulse sliding window in combination with the water film thickness change rate, and determine the pulse incidence rate in combination with the trigger times of the grid fault event; In response to the input time-domain features, the number of discharge sources, the thermal stress suppressed acoustic energy, and the pulse incidence rate, establish a feature vector in combination with the phase window ratio, and detect the fault state of the distribution network according to the discrimination model.

2. The method for detecting faults in a distribution network line based on data analysis according to claim 1, wherein When the amplitude of the current signal is greater than the trigger threshold and the interval from the previous trigger is greater than or equal to the minimum sampling point interval, record a pulse, and the trigger sampling sequence number is ; Extract the pulse time-domain window: ； 、 Respectively represent the number of sampling points on both sides of the window width, Indicates the The intercepted signal within the time domain window where the trigger event is located, represents the pulse current time domain sequence of the original HFCT sampling; Establish a superposition model: ； Indicates the peak amplitude of the pure discharge of the nth pulse, Indicates the exponential decay constant, Indicates the transient reference voltage on the insulator surface, Indicates the time corresponding to the nth sampling point within the pulse window, Indicates the instantaneous equivalent capacitance corresponding to the time of the nth sampling point, Indicates the time corresponding to the nth sampling point The instantaneous capacitance between the lower conductor and the ground, Indicates the line voltage value at the nth sampling point.

3. The method for detecting faults in distribution network lines based on data analysis according to claim 1, wherein Introduce a time-varying capacitance model to compensate for the trailing effect of the water film dynamic process on the capacitance current, so as to extract the pure discharge peak amplitude and the exponential decay constant. Among them, the time-varying capacitance model is specifically: ； Indicates the water film capacitance at a moment, indicates the reference capacitance when the pollution layer is unsaturated, indicates the additional capacitance after the water film is formed, indicates the water film evaporation rate constant, indicates a real variable, Indicates the th trigger moment.

4. The method for detecting faults in distribution network lines based on data analysis according to claim 2, wherein Water film trailing coefficient , which measures the degree of distortion of the pure discharge signal by the water film coupling: ； Wind vibration suppression coefficient , measuring the interference degree of conductor-ground coupling on the pure discharge signal: ； Based on the water film trailing coefficient and the wind vibration suppression coefficient to determine surface partial discharge or line arc fault.

5. The method for detecting faults in a distribution network line based on data analysis according to claim 1, wherein The method for obtaining the thermal stress suppressed acoustic energy is: Temperature gradient Calculation: ； Indicates the time interval of environmental quantity sampling, Indicates the environmental temperature corresponding to the th pulse trigger time; Calculate the ultrasonic energy during the th pulse: ； represents the acoustic signal value at the Define the coefficient of thermal stress acoustic suppression ; represents the reference temperature, represents the temperature sensitivity coefficient; If the following conditions are met and are respectively greater than their corresponding thresholds, it is determined that the ultrasonic energy comes from the thermal stress crack noise, and the thermal stress suppression acoustic energy is obtained by suppression: , represents the thermal stress threshold, represents the suppression threshold.

6. The method for detecting faults in a distribution network line based on data analysis according to claim 2, wherein The method for determining the pulse incidence rate is: Count the number of trigger times within a time period ; Indicates the index of the trigger event, Indicates the sequence number of the discrete sampling point corresponding to the th pulse trigger, Indicates the sampling rate of the high-frequency current transformer, Indicates the dynamic sliding window length corresponding to the th trigger; Represents the counting operation of the number of set elements; Calculate the pulse incidence rate according to the ratio of the statistical trigger times to the length of the dynamic sliding window.

7. The method for detecting faults in a distribution network line based on data analysis according to claim 1, characterized in that, The method for the discrimination model to detect the fault state of the distribution network is: Construct a logistic regression model and input the feature vector to output the arc discharge probability: Construct a cross-entropy loss function; Calculate the arc discharge probability according to the feature vector obtained each time; If it is greater than the probability threshold, it is determined as a cable arc fault, otherwise it is a surface partial discharge.

8. A distribution network line fault detection system, characterized in that, It is implemented according to the distribution network line fault detection method based on data analysis described in any one of claims 1-7.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method described in any one of claims 1 to 7.

Images