An environment-friendly gas insulated ring main unit operation fault diagnosis method

Abstract

The application discloses an environment-friendly gas insulated ring main unit operation fault diagnosis method and relates to the technical field of insulation ring main units. The method comprises the following steps: collecting partial discharge data of a metal closed gas chamber, extracting a time-frequency complex matrix, a repetition rate density and a phase angle feature; combining frequency energy center analysis and high-frequency energy proportion calculation to calculate frequency energy comprehensive characteristic values; positioning a discharge signal position sequence by using an inverse distance weighting method, fusing the position sequence, the frequency energy comprehensive characteristic values, the repetition rate density and the phase angle feature to map into a discharge signal pseudo-color spectrum and mark; inputting the marked spectrum into a convolutional neural network for training, collecting air pressure disturbance data by using a micro pressure difference sensor and extracting features to input into a long short-term memory network for training; collecting discharge pseudo-color spectrum and air pressure disturbance features in real time to input into the trained convolutional neural network and long short-term memory network model, outputting discharge signal categories and air pressure disturbance feature prediction values and calculating insulation ring main unit risk values.

Description

Technical Field

[0001] This invention relates to the field of insulated ring main units, specifically to a method for diagnosing operational faults in environmentally friendly gas-insulated ring main units. Background Technology

[0002] In the process of intelligent development of power distribution networks, environmentally friendly gas-insulated ring main units have become one of the core equipment in urban power grid construction and renovation due to their advantages such as green environmental protection, compact structure, and high reliability. Their safe and stable operation is crucial to the power supply reliability of the distribution network, and accurate fault diagnosis is a key link in ensuring equipment operation. Currently, the industry's fault diagnosis of insulated ring main units largely relies on the monitoring of single physical quantities, such as the amplitude and frequency characteristics of discharge signals or independent analysis of gas pressure disturbances.

[0003] However, existing technologies only analyze the frequency characteristics of discharge signals to determine the discharge type, but ignore the location of the discharge signal in the metal-sealed gas chamber. The degree of danger of the discharge signal varies at different locations, which leads to inaccurate fault diagnosis. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for diagnosing operational faults in environmentally friendly gas-insulated ring main units, thereby resolving the problems existing in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit, comprising the following steps:

[0006] Step S1: Collect partial discharge data from the metal-sealed gas chamber to obtain a discharge signal dataset;

[0007] Step S2: By extracting features from the discharge signal dataset, discharge signal feature data is obtained. The discharge signal feature data includes: time-frequency complex matrix, repetition rate density, and phase angle features.

[0008] Step S3: Obtain the frequency center energy ratio by performing frequency energy center analysis on the time-frequency complex matrix; obtain the high-frequency energy proportion by performing high-frequency energy analysis on the time-frequency complex matrix;

[0009] Step S4: Combine the frequency center energy ratio and the high-frequency energy ratio to calculate the comprehensive frequency energy characteristic value; based on the discharge signal dataset, obtain the discharge signal position sequence; perform two-dimensional matrix mapping and annotation on the discharge signal position sequence, comprehensive frequency energy characteristic value, repetition rate density and phase angle characteristics to obtain the annotated discharge signal pseudo-color spectrum.

[0010] Step S5: Input the labeled pseudo-color spectrum of the discharge signal into the convolutional neural network model for training to obtain the trained convolutional neural network model; collect the pressure disturbance data of the metal sealed air chamber, extract features to obtain the pressure disturbance feature data, and input it into the long short-term memory network model for training to obtain the trained long short-term memory network model.

[0011] Step S6: Collect the pseudo-color spectrum of the discharge signal and the characteristic data of the air pressure disturbance in real time, and input them into the trained convolutional neural network model and the trained long short-term memory network model respectively. The corresponding outputs are the predicted values ​​of the discharge signal category and the air pressure disturbance characteristics, which are then combined to calculate the risk value of the insulated ring main unit. Based on the risk value of the insulated ring main unit, the fault diagnosis of the insulated ring main unit is realized.

[0012] Preferably, the step of extracting features from the discharge signal dataset to obtain discharge signal feature data includes the following specific steps:

[0013] Frequency conversion is performed using short-time Fourier transform to obtain the time-frequency complex matrix of the discharge signal:

[0014]

[0015] in, Let L be the short-time Fourier transform output value of the m-th overlapping frame signal at frequency point k, and L be the length of each overlapping frame, 64. <L 4096, Let k be the m-th overlapping frame signal of the discharge signal, where m is the index of the m-th overlapping frame and k is the frequency index. K is the window function, K is the frame shift, and n is the index of the sampling point within the overlapping frame.

[0016] Calculate the repetition density for each time window:

[0017]

[0018] in, For time repetition rate density, This represents the total number of discharge pulses within the time window. For the Dirac function, Let j be the total length of the j-th time window. = T represents the total acquisition time, and J represents the total number of time windows;

[0019] Calculate the phase dispersion for each time window:

[0020]

[0021] in, For phase dispersion, This represents the total number of discharge pulses within the time window. , For the first The power frequency phase corresponding to the time of each discharge pulse occurrence , This is the index of the discharge pulse.

[0022] Preferably, the step of obtaining the frequency center energy ratio by performing frequency energy center analysis on the time-frequency complex matrix includes the following steps:

[0023] By performing frequency energy center analysis on the time-frequency complex matrix within each time window, the frequency center energy ratio is obtained:

[0024]

[0025] in, Here, M represents the frequency center energy ratio, M is the total number of overlapping frames, and m is the index of the m-th overlapping frame. Let B be the center frequency index of the m-th overlapping frame, B be the center bandwidth, L be the length of each overlapping frame, and k be the frequency index. This represents the short-time Fourier transform of the m-th overlapping frame signal at frequency point k.

[0026] Preferably, the high-frequency energy percentage is:

[0027]

[0028] in, Here, M represents the proportion of high-frequency energy, m is the total number of overlapping frames, m is the index of the m-th overlapping frame, L is the length of each overlapping frame, and k is the frequency index. This represents the short-time Fourier transform of the m-th overlapping frame signal at frequency point k. This is the frequency index corresponding to the high-frequency threshold.

[0029] Preferably, the calculation of the comprehensive frequency energy characteristic value by combining the frequency center energy ratio and the high-frequency energy ratio includes the following steps:

[0030] By combining the frequency center energy ratio and high-frequency energy proportion within each time window, the comprehensive frequency energy characteristic value is calculated:

[0031]

[0032] in, This is the comprehensive characteristic value of frequency and energy. The frequency center energy ratio, This represents the proportion of high-frequency energy.

[0033] Preferably, obtaining the discharge signal location sequence based on the discharge signal dataset includes the following specific steps:

[0034] The discharge signal location sequence was calculated by performing location analysis on the discharge signal dataset within each time window using the inverse distance weighting method.

[0035]

[0036] in, Indicates within the time window Intrinsic coordinates ( The discharge probability under ( ) is given, where N is the total number of sensors, and N>3. It is the first It is the first The maximum amplitude of the received discharge signal, It is a metal-enclosed air chamber space area. It is an indicator function, indicating that only if the point... The value is 1 when the chamber is enclosed in a metal cavity, and 0 otherwise.

[0037] The discharge probability is compared with a preset discharge probability threshold. If it is greater than the preset threshold, it is marked as the cylindrical coordinate of the discharge signal. The cylindrical coordinates of all the filtered discharge signals are combined to obtain the discharge signal position sequence within each time window. The discharge signal position sequence of the nth time window is then obtained. The cylindrical coordinates of the discharge signal are .

[0038] Preferably, the step of obtaining the labeled pseudo-color spectrum of the discharge signal by performing two-dimensional matrix mapping and annotation on the discharge signal position sequence, frequency energy comprehensive feature value, repetition rate density and phase angle feature includes the following specific steps:

[0039] Map all points on the cylindrical coordinate system of the sealed metal chamber to two-dimensional coordinates to construct a pseudo-color spectrum of the discharge signal:

[0040]

[0041] in, For the first The two-dimensional horizontal axis of the pseudo-color spectrum of a discharge signal. For the first The azimuth angle of the cylindrical coordinate system of the discharge signal. The resolution of the two-dimensional matrix of the pseudo-color spectrum of the discharge signal;

[0042]

[0043] in, For the first The two-dimensional vertical axis of the pseudo-color spectrum of the discharge signal. For the first The height of the discharge signal cylindrical coordinate system, where H is the total height of the metal-sealed gas chamber. The resolution of the two-dimensional matrix of the pseudo-color spectrum of the discharge signal;

[0044] In the cylindrical coordinate system of each discharge signal The ratio of the total radius R of the metal-enclosed gas chamber to the discharge signal pseudo-color spectrum is used as the first value in the spectrum. Two-dimensional coordinates of a discharge signal ( , Mapping of the red channel:

[0045]

[0046] in, For the first The mapping of the discharge signal in the red channel, For the first The radius of the discharge signal in the cylindrical coordinate system, R is the total radius of the metal-enclosed gas chamber, and clip() is the clip function;

[0047] The frequency-energy comprehensive characteristic value of each discharge signal and the total number of times the discharge position sequence appears in all time windows within the total acquisition time T are used as the first value in the pseudocolor spectrum of the discharge signal. Two-dimensional coordinates of a discharge signal ( , Mapping of the green channel:

[0048]

[0049] in, For the first The mapping of a discharge signal in the green channel of the discharge signal pseudo-color spectrum. For the first The frequency-energy comprehensive characteristic value of each discharge signal, where clip() is the clip function. For the first The total number of times a discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T. This represents the maximum number of times the discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T;

[0050] The phase dispersion, repetition density, and total number of times the discharge position sequence appears in all time windows within the total acquisition time T for each discharge signal are used as the index of the discharge signal pseudocolor spectrum. Two-dimensional coordinates of a discharge signal ( , Mapping of the green channel:

[0051]

[0052] in, For the first The mapping of a discharge signal in the blue channel of the discharge signal pseudocolor spectrum. For the first Phase dispersion of each discharge signal, For the first The repetition rate density of each discharge signal For the first The maximum number of pulses of a discharge signal within a time window, where clip() is the clip function. For the first The total number of times a discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T. This represents the maximum number of times the discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T;

[0053] The pseudo-color spectrum of the discharge signal is annotated using the discharge signal position sequence, and finally an annotated pseudo-color spectrum of the discharge signal is obtained.

[0054] Preferably, the real-time collection of the pseudo-color spectrum of the discharge signal and the pressure disturbance feature data, and the input of these data into the trained convolutional neural network model and the trained long short-term memory network model, respectively, to output the discharge signal category and the predicted value of the pressure disturbance feature, includes the following specific steps:

[0055] The pseudo-color spectrum of the discharge signal is collected in real time within a total time window of T, input into a trained convolutional neural network model, and the discharge signal category is output.

[0056]

[0057] in, For the prediction of the convolutional neural network model The probability of each label category. For the convolutional neural network model to the first The original predicted values ​​for each label are obtained from the output of the fully connected layer;

[0058] The label with the highest predicted probability among all labels in the neural network model is selected as the label for the pseudo-color spectrum of the discharge signal collected in real time. , indicating the first One tag;

[0059] The system collects and normalizes a sequence of air pressure disturbance feature data over a total time window of T, then inputs it into a trained long short-term memory network model to output predicted air pressure disturbance features.

[0060]

[0061] in, This represents the predicted values ​​of amplitude change, steepness slope, and response lag time over J+1 time windows. This is the weight matrix of the fully connected layer in a Long Short-Term Memory (LSTM) network model. This represents the hidden states of the Long Short-Term Memory network model at time T. This is the bias term for the fully connected layer in a Long Short-Term Memory (LSTM) network model.

[0062] Preferably, the risk value of the insulated ring main unit is:

[0063]

[0064] Among them, FX represents the risk value of the insulated ring main unit. Labels for the pseudo-color spectra of discharge signals collected in real time. This represents the predicted amplitude change over the J+1 time window.

[0065] Preferably, the step of diagnosing faults in the insulated ring main unit based on its risk value includes the following specific steps:

[0066] The risk value of the insulated ring main unit is compared with a preset threshold. When the risk value of the insulated ring main unit is greater than the preset threshold, it indicates that the current discharge characteristics and air pressure disturbance characteristics have exceeded the normal operating range, and there is a high risk of insulation failure. This triggers the early warning mechanism, which provides a visual interface prompt and issues an alarm to the operation and maintenance personnel. At the same time, it combines the discharge signal type to accurately locate the potential fault area and provide clear guidance for subsequent maintenance decisions.

[0067] Beneficial effects:

[0068] This invention provides a method for diagnosing operational faults in environmentally friendly gas-insulated ring main units, involving machine learning and deep learning technologies, which has the following beneficial effects:

[0069] (1) By combining the frequency center energy ratio and the high-frequency energy ratio to calculate the comprehensive characteristic value of frequency energy, the energy distribution characteristics and frequency component ratio of the discharge signal can be effectively characterized. The frequency center energy ratio reflects the degree of concentration of discharge energy near the center frequency, while the high-frequency energy ratio reflects the contribution of high-frequency energy to the overall energy. The fusion of the two can sensitively capture the differential effects of discharge types such as corona discharge and surface discharge on frequency characteristics, providing key parameters for the subsequent construction of a pseudo-color spectrum of the discharge signal containing energy characteristics, enhancing the distinguishability of the feature space, and thus improving the accuracy of the convolutional neural network model in identifying discharge types.

[0070] (2) Generating a pseudo-color spectrum of the discharge signal by mapping the discharge signal position sequence, frequency-energy comprehensive feature value, repetition rate density, and phase angle features into a two-dimensional matrix is ​​a key step in transforming multi-dimensional abstract features into a visualized image. By mapping the spatial coordinates of the gas chamber to two-dimensional planar coordinates, determining the spatial position by the discharge position, and using the frequency-energy comprehensive feature value, repetition rate density, etc., as color channel parameters, a multi-dimensional feature spectrum containing spatial distribution, energy intensity, time frequency, and phase characteristics can be constructed. This mapping method not only realizes the spatial visualization of discharge features, but also integrates multi-physical quantity information through color coding, enabling the convolutional neural network to automatically extract the texture, shape, and color distribution features in the spectrum, effectively overcoming the shortcomings of traditional feature vector analysis in utilizing spatial correlation information, and providing richer input dimensions for discharge pattern recognition.

[0071] (3) By combining the discharge signal category and the predicted value of the gas pressure disturbance characteristics, the risk value of the insulated ring main unit is calculated, realizing a multi-dimensional quantitative assessment of the equipment's operating status. The discharge location category reflects the spatial distribution characteristics of the discharge, while the predicted value of the gas pressure disturbance characteristics reflects the response characteristics of the gas state to the discharge. The fusion of the two not only covers the spatial migration law of discharge development but also includes the temporal information of changes in gas insulation performance. By establishing a risk value calculation model, the dynamic coupling relationship between spatial and temporal characteristics can be transformed into specific values, which can be compared with preset thresholds to achieve graded early warning of faults. This comprehensive assessment method breaks through the limitations of monitoring a single physical quantity, can more comprehensively reflect the insulation degradation process, provide scientific decision-making basis for operation and maintenance personnel, and improve the level of precision in equipment condition management. Attached Figure Description

[0072] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0073] Figure 1This is a flowchart of the steps for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit proposed in this invention.

[0074] Figure 2 This is a step-by-step diagram of a method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit proposed in this invention. Detailed Implementation

[0075] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] Please see Figure 1-2 The present invention provides a technical solution: a method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit.

[0077] Step S1: Collect partial discharge data from the metal-sealed gas chamber to obtain a discharge signal dataset.

[0078] Based on the air chamber structure (e.g., cylindrical) and monitoring requirements, various types of sensors are deployed on the surface or inside the air chamber at key locations (e.g., top, sides, bottom, near insulators), including ultra-high frequency (UHF) sensors (capturing 300MHz~3GHz electromagnetic waves), high frequency current sensors (HFCT, detecting 1MHz~30MHz grounding current), and ultrasonic sensors (AE, monitoring 20kHz~200kHz mechanical sound waves), forming a spatially covering sensor network.

[0079] All sensors are pre-calibrated, including sensitivity calibration, time delay measurement, and spatial coordinate registration (e.g., determining the sensor's three-dimensional position via laser ranging), and connected to a high-precision synchronization clock (e.g., a GPS synchronization module) to ensure that the sampling timestamp error of each sensor is <1μs. A multi-channel data acquisition system is built, with sampling parameters set as follows: UHF sensor sampling rate ≥1GHz, HFCT and AE sensor sampling rate ≥10MHz, continuous or triggered acquisition mode (trigger threshold can be set to 2-3 times the background noise). Under air chamber operation conditions (e.g., normal power supply, load switching, etc.), the raw electrical or acoustic signals output by the sensors are continuously acquired, and environmental parameters (temperature, humidity, air pressure) and air chamber operating parameters (voltage, current, switch position) are recorded simultaneously.

[0080] During the acquisition process, the raw signals are displayed in real time with waveforms and anomaly markers to prevent data interruption or missed outliers. Data is stored in files based on sensor type and channel. Each data file contains a timestamp sequence, raw waveform data (such as the sampling point sequence of time-domain voltage / current / sound pressure signals), sensor ID, spatial coordinates, and associated metadata (environmental and operational parameters). The storage format uses an efficient binary format (such as HDF5 or MAT files) for subsequent analysis. After acquisition, the raw dataset is verified for integrity, and segments with sampling anomalies or synchronization failures are removed, resulting in a discharge signal dataset without feature extraction.

[0081] Step S2: By extracting features from the discharge signal dataset, discharge signal feature data is obtained. The discharge signal feature data includes: time-frequency complex matrix, repetition rate density, and phase angle features.

[0082] The total data collection time is T, and the total time T is divided into J time windows. , , ,..., ,..., ],in Let j be the j-th time window, and let the sampling frequency within each time window be . The number of sampling points is The discharge signal is S[n].

[0083] By extracting features from the discharge signal dataset within each time window, discharge signal feature data is obtained, which includes: time-frequency complex matrix, repetition rate density, and phase angle features.

[0084] The discharge signal is divided into M overlapping frames, each with a length of L points and a frame shift of K points. The discrete signal of the m-th overlapping frame is: , This is the m-th overlapping frame signal of the discharge signal. This is the original discharge signal. For example, the Hanning window is a window function.

[0085] It should be noted that the discharge signal data within each time window is the average of all sensors within that time window.

[0086] Frequency conversion is performed using short-time Fourier transform to obtain the time-frequency complex matrix of the discharge signal:

[0087]

[0088] in, Let L be the short-time Fourier transform output value of the m-th overlapping frame signal at frequency point k, and L be the length of each overlapping frame, 64. <L 4096, Let k be the m-th overlapping frame signal of the discharge signal, where m is the index of the m-th overlapping frame and k is the frequency index. K is the window function, K is the frame shift, and n is the index of the sampling point within the overlapping frame.

[0089] Calculate the repetition density for each time window:

[0090]

[0091] in, For time repetition rate density, This represents the total number of discharge pulses within the time window. For the Dirac function, Let j be the total length of the j-th time window. = T represents the total acquisition time, and J represents the total number of time windows.

[0092] It should be noted that the total number of discharge pulses within the time window... This refers to dividing the total data collection time T into J time windows, and then processing each time window (e.g., the j-th time window) accordingly. This is the cumulative number of discharge pulses detected after signal processing from discharge signals collected by various types of sensors (such as UHF, HF current, and ultrasonic sensors). This statistical count is based on the discharge signal dataset within each time window (i.e., the average of all sensor data within that window), reflecting the frequency of partial discharge activity inside the metal-sealed gas chamber during that period.

[0093] Calculate the phase dispersion for each time window:

[0094]

[0095] in, For phase dispersion, This represents the total number of discharge pulses within the time window. For the first The power frequency phase corresponding to the time of each discharge pulse occurrence , This is the index of the discharge pulse.

[0096] Step S3: By performing frequency energy center analysis on the time-frequency complex matrix, the frequency center energy ratio is obtained; by performing high-frequency energy analysis on the time-frequency complex matrix, the high-frequency energy proportion is obtained.

[0097] By performing frequency energy center analysis on the time-frequency complex matrix within each time window, the frequency center energy ratio is obtained:

[0098]

[0099] in, Here, M represents the frequency center energy ratio, M is the total number of overlapping frames, and m is the index of the m-th overlapping frame. Let B be the center frequency index of the m-th overlapping frame, B be the center bandwidth, L be the length of each overlapping frame, and k be the frequency index. This represents the short-time Fourier transform of the m-th overlapping frame signal at frequency point k.

[0100] By performing high-frequency energy analysis on the aforementioned time-frequency complex matrix, the proportion of high-frequency energy is obtained:

[0101]

[0102] in, Here, M represents the proportion of high-frequency energy, m is the total number of overlapping frames, m is the index of the m-th overlapping frame, L is the length of each overlapping frame, and k is the frequency index. This represents the short-time Fourier transform of the m-th overlapping frame signal at frequency point k. This is the frequency index corresponding to the high-frequency threshold.

[0103] Step S4: Combine the frequency center energy ratio and the high-frequency energy ratio to calculate the comprehensive frequency energy characteristic value; based on the discharge signal dataset, obtain the discharge signal position sequence; by performing two-dimensional matrix mapping and annotation on the discharge signal position sequence, the comprehensive frequency energy characteristic value, the repetition rate density, and the phase angle characteristics, obtain the annotated discharge signal pseudo-color spectrum.

[0104] By combining the frequency center energy ratio and high-frequency energy proportion within each time window, the comprehensive frequency energy characteristic value is calculated:

[0105]

[0106] in, This is the comprehensive characteristic value of frequency and energy. The frequency center energy ratio, This represents the proportion of high-frequency energy.

[0107] The discharge signal location sequence was calculated by performing location analysis on the discharge signal dataset within each time window using the inverse distance weighting method.

[0108]

[0109] in, Indicates within the time window Intrinsic coordinates ( The discharge probability under ( ) is given, where N is the total number of sensors, and N>3. It is the first It is the first The maximum amplitude of the received discharge signal, It is a metal-enclosed air chamber space area. It is an indicator function, indicating that only if the point... The value is 1 when the chamber is enclosed in a metal chamber, and 0 otherwise.

[0110] It should be noted that most metal enclosed air chambers are cylindrical, so a cylindrical coordinate system is used. The center of the circle is defined as the geometric center of the cross-section of the air chamber (i.e., the circular plane perpendicular to the axis), which is usually located at the center point of the bottom circle of the metal enclosed air chamber, and this center point coincides with the axis (z-axis) of the metal enclosed air chamber. r (radial coordinate): represents the distance from the measuring point to the center of the circle (i.e., the radius of the measuring point on the cross-section). The center of the circle is the position where r=0. The maximum radial distance of the inner wall of the metal enclosed air chamber is the total radius R (i.e., r∈[0,R]). (Azimuth): The angle of counterclockwise rotation around the center (axis) of a circle, with a fixed radial line passing through the center (such as the central axis of the air chamber) as the 0° reference. ∈[0°, 360°). z (height coordinates): With the center of the bottom surface of the metal sealed air chamber as the origin, the z-axis is the axis passing through the center and perpendicular to the bottom surface, with upward as the positive direction. The total height of the metal sealed air chamber is H (i.e., z∈[0, H]).

[0111] The discharge probability is compared with a preset discharge probability threshold (e.g., 0.8). , The maximum discharge probability of all coordinates is compared. If it is greater than a preset threshold, it is marked as the cylindrical coordinate of the discharge signal. The cylindrical coordinates of all filtered discharge signals are combined to obtain the discharge signal position sequence within each time window. The position of the discharge signal in the discharge signal position sequence is then determined. The cylindrical coordinates of the discharge signals are ( ).

[0112] Map all points on the cylindrical coordinate system of the sealed metal chamber to two-dimensional coordinates to construct a pseudo-color spectrum of the discharge signal:

[0113]

[0114] in, For the first The two-dimensional horizontal axis of the pseudo-color spectrum of a discharge signal. For the first The azimuth angle of the cylindrical coordinate system of the discharge signal. The resolution of the two-dimensional matrix of the pseudo-color spectrum of the discharge signal.

[0115]

[0116] in, For the first The two-dimensional vertical axis of the pseudo-color spectrum of the discharge signal. For the first The height of the discharge signal cylindrical coordinate system, where H is the total height of the metal-sealed gas chamber. The resolution of the two-dimensional matrix of the pseudo-color spectrum of the discharge signal.

[0117] Then, the two-dimensional coordinates corresponding to the cylindrical coordinates of each discharge signal in the discharge signal position sequence within each time window are ( , ), and count the number The total number of times the discharge signal appears in the discharge location sequence across all time windows within the total acquisition time T. The discharge signals in the discharge location sequence within all time windows within the total acquisition time T are mapped to the color channels of the discharge signal pseudo-color spectrum.

[0118] For the two-dimensional coordinates of the discharge signal, the cylindrical coordinates of each discharge signal are... The ratio of the total radius R of the metal-sealed gas chamber is used as the mapping of the red channel in the pseudo-color spectrum of the discharge signal. The total number of times the frequency-energy comprehensive characteristic value of each discharge signal appears in the discharge position sequence within all time windows within the total acquisition time T is used as the mapping of the green channel in the pseudo-color spectrum of the discharge signal. The phase dispersion, repetition density, and total number of times the discharge position sequence appears in all time windows within the total acquisition time T of each discharge signal are used as the mapping of the blue channel in the pseudo-color spectrum of the discharge signal.

[0119] In the cylindrical coordinate system of each discharge signal The ratio of the total radius R of the metal-enclosed gas chamber to the discharge signal pseudo-color spectrum is used as the first value in the spectrum. Two-dimensional coordinates of a discharge signal ( , Mapping of the red channel:

[0120]

[0121] in, For the first The mapping of the discharge signal in the red channel, For the first The radius of the discharge signal in the cylindrical coordinate system, R is the total radius of the metal-sealed gas chamber, and clip() is the clip function. For the first The total number of times a discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T.

[0122] The frequency-energy comprehensive characteristic value of each discharge signal and the total number of times the discharge position sequence appears in all time windows within the total acquisition time T are used as the first value in the pseudocolor spectrum of the discharge signal. Two-dimensional coordinates of a discharge signal ( , Mapping of the green channel:

[0123]

[0124] in, For the first The mapping of a discharge signal in the green channel of the discharge signal pseudo-color spectrum. For the first The frequency-energy comprehensive characteristic value of each discharge signal, where clip() is the clip function. For the first The total number of times a discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T. This represents the maximum number of times the discharge signal appears in the discharge position sequence across all time windows of the total acquisition time T.

[0125] The phase dispersion, repetition density, and total number of times the discharge position sequence appears in all time windows within the total acquisition time T for each discharge signal are used as the index of the discharge signal pseudocolor spectrum. Two-dimensional coordinates of a discharge signal ( , Mapping of the green channel:

[0126]

[0127] in, For the first The mapping of a discharge signal in the blue channel of the discharge signal pseudocolor spectrum. For the first Phase dispersion of each discharge signal, For the first The repetition rate density of each discharge signal For the first The maximum number of pulses of a discharge signal within a time window, where clip() is the clip function. For the first The total number of times a discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T. This represents the maximum number of times the discharge signal appears in the discharge position sequence across all time windows of the total acquisition time T.

[0128] It should be noted that the green channel maps to the comprehensive energy characteristic value, with a value range of 0-255. The higher the green channel mapping value, the stronger the discharge energy. The blue channel maps to the phase dispersion and repetition rate density, with a value range of 0-255. The higher the blue channel mapping value, the more unstable and frequent the discharge.

[0129] The pseudo-color spectrum of the discharge signal is annotated by the position sequence of the discharge signal to obtain the annotated pseudo-color spectrum of the discharge signal.

[0130] It should be noted that the labeled pseudo-color spectrum of the discharge signal is obtained by calculating the pixel mean of the green channel of all discharge signals in the pseudo-color spectrum. and the average pixel value of the blue channel The discharge intensity levels are classified as low, medium, and high intensity based on the average pixel value of the green channel; the discharge characteristics are also classified as stable and unstable discharge based on the average pixel value of the green channel. The three-dimensional cylindrical coordinates of the metal-enclosed gas chamber are divided into bottom, middle, and top regions according to height, and this is combined with the mapping of each discharge signal in the red channel. To determine if a region is along a surface, the number of discharge signals in each region is counted. A threshold for the number of discharges is set. When the number of discharges in a region exceeds the threshold, it is marked as a single-region discharge (e.g., bottom discharge, middle surface discharge). If the number of discharges in two or more adjacent regions both exceed the threshold, they are combined and marked as cross-region discharge (e.g., bottom-middle discharge, full-region discharge). For example, when the pseudo-color spectrum of the discharge signal is the bottom region and the average pixel value of the green channel is... Average pixel value for high intensity and blue channel If the discharge level is stable, then the pseudo-color spectrum of the discharge signal is marked as "high-intensity bottom stable discharge".

[0131] Step S5: Input the labeled pseudo-color spectrum of the discharge signal into the convolutional neural network model for training to obtain the trained convolutional neural network model; collect the pressure disturbance data of the metal sealed air chamber and extract its features to obtain the pressure disturbance feature data, and input it into the long short-term memory network model for training to obtain the trained long short-term memory network model.

[0132] Several labeled pseudo-color spectra of discharge signals were collected as training data. These were preprocessed, including normalization, to unify the scale before being input into a convolutional neural network model for training. During training, the model automatically extracted spatial features such as color, texture, and shape from the pseudo-color spectra of the discharge signals through convolutional layers. Pooling layers reduced the feature dimensionality and enhanced translation invariance. Fully connected layers then mapped the extracted features to specific discharge types or location categories. Backpropagation was used to optimize the model parameters, and the process was iterated until the loss function converged, ultimately yielding the trained convolutional neural network model.

[0133] The structure of the convolutional neural network model includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer.

[0134] A convolutional neural network structure consists of an input layer, four stages, one fully connected layer, and a softmax output layer.

[0135] Input layer: Receives the preprocessed discharge signal pseudo-color spectrum, with a size adapted to the image size and 3 channels corresponding to the RGB three channels.

[0136] Stage 1 consists of two convolutional layers and one pooling layer. Each convolutional layer has a depth of 32, and the activation function is the Sigmoid() function. The pooling layer uses max pooling for resampling. Stage 2 consists of two convolutional layers and one pooling layer. Each convolutional layer has a depth of 64, and the activation function is the Sigmoid() function. The pooling layer uses max pooling for resampling. Stage 3 consists of two convolutional layers and one pooling layer. Each convolutional layer has a depth of 128, and the activation function is the Sigmoid() function.

[0137] The pooling layer uses max pooling for resampling; Stage 4 consists of three convolutional layers and one pooling layer. Each convolutional layer has a depth of 256 and uses the Sigmoid() activation function. The pooling layer uses max pooling for resampling.

[0138] After acquiring the air pressure data of the metal-sealed air chamber using a micro differential pressure sensor, the data is divided into time windows (such as fixed duration or sliding windows) identical to those of the discharge signal. A time window matching mechanism is added to ensure that the air pressure disturbance data within each time window is completely synchronized with the discharge signal in terms of time range and timing. Within each time window, the following features are extracted from the air pressure disturbance data: amplitude variation, steepness slope, and response lag time of the pressure disturbance waveform.

[0139] It should be noted that the amplitude change of the pressure disturbance waveform, calculated as the difference between the maximum and minimum amplitudes of the air pressure waveform within the time window, reflects the intensity range of the air pressure fluctuation.

[0140] It should be noted that the steepness slope is obtained by analyzing the slope of the rising or falling edge of the air pressure waveform to extract the rate of change of the waveform. For example, by measuring the time interval from the start point of the disturbance to the peak point, the amplitude change per unit time (i.e., the slope) is calculated. The larger the slope, the more severe the disturbance.

[0141] It should be noted that the response lag time, which determines the time difference between the trigger moment of the discharge signal and the moment when the pressure disturbance waveform begins to change significantly, is used to locate the lag time of the pressure disturbance relative to the discharge event through threshold detection or waveform abrupt change point identification. This is used to analyze the delay characteristics of the gas pressure response. All of these features are extracted based on a time window synchronized with the discharge signal, forming pressure disturbance feature data containing parameters such as amplitude variation range, slope value, and lag time. This provides multi-dimensional, time-aligned information for subsequent joint analysis combining discharge signal features.

[0142] The pressure disturbance data collected and synchronized with the pseudo-color spectra of the discharge signal are combined into disturbance features, such as amplitude variation, steep slope, and response lag time. After normalization, these features are input into the long short-term memory network model for training. By capturing the temporal dependence of the pressure disturbance data sequence, the correlation between the discharge mode and pressure change is learned. The prediction error is minimized through the backpropagation algorithm, and finally, the trained long short-term memory network model is obtained.

[0143] A Long Short-Term Memory (LSTM) network model consists of an input layer, a feature extraction layer, and a fully connected output layer. Input layer design...

[0144] The input data is a normalized time-series feature vector of pressure disturbances, including amplitude change rate, steep slope, and response lag time. Feature extraction layer: A bidirectional LSTM (Bi-LSTM) structure is used to capture the bidirectional dependencies of the pressure disturbance sequence through forward and backward propagation. The specific configuration is as follows: First Bi-LSTM layer: 128 memory units, return_sequences=True to retain the output of each time step, facilitating subsequent network stacking to capture multi-level time-series features. Second Bi-LSTM layer: 64 memory units, return_sequences=False to output only the hidden state of the last time step, achieving feature compression. Dropout mechanism: Dropout rates of 0.2 and 0.3 are added to the two Bi-LSTM layers respectively to prevent overfitting and enhance the model's generalization ability. Fully connected output layer: Two fully connected network layers map the time-series features extracted by the LSTM to the predicted values ​​of the pressure disturbances: The intermediate fully connected layer contains 32 neurons and uses the ReLU activation function to further fuse the non-linear correlations of the time-series features. Output layer: Contains 3 neurons (corresponding to 3D air pressure perturbation features), and directly outputs the predicted value using a linear activation function.

[0145] Step S6: Collect the pseudo-color spectrum of the discharge signal and the characteristic data of the air pressure disturbance in real time, and input them into the trained convolutional neural network model and the trained long short-term memory network model respectively. The corresponding outputs are the predicted values ​​of the discharge signal category and the air pressure disturbance characteristics, which are then combined to calculate the risk value of the insulated ring main unit. Based on the risk value of the insulated ring main unit, the fault diagnosis of the insulated ring main unit is realized.

[0146] The pseudo-color spectrum of the discharge signal and the characteristic data of air pressure disturbance are collected in real time and input into the trained convolutional neural network model and the trained long short-term memory network model, respectively, and the corresponding outputs are the discharge signal category and the predicted value of air pressure disturbance characteristics.

[0147] The pseudo-color spectrum of the discharge signal is collected in real time within a total time window of T, input into a trained convolutional neural network model, and the discharge signal category is output.

[0148]

[0149] in, For the prediction of the convolutional neural network model The probability of each label category. For the convolutional neural network model to the first The original predicted values ​​of each label are obtained from the output of the fully connected layer.

[0150] The label with the highest predicted probability among all labels in the neural network model is selected as the label for the pseudo-color spectrum of the discharge signal collected in real time. , indicating the first A tag.

[0151] The system collects and normalizes a sequence of air pressure disturbance feature data over a total time window of T, then inputs it into a trained long short-term memory network model to output predicted air pressure disturbance features.

[0152]

[0153] in, This represents the predicted values ​​of amplitude change, steepness slope, and response lag time over J+1 time windows. This is the weight matrix of the fully connected layer in a Long Short-Term Memory (LSTM) network model. This represents the hidden states of the Long Short-Term Memory network model at time T. This is the bias term for the fully connected layer in a Long Short-Term Memory (LSTM) network model.

[0154] By combining the predicted values ​​of discharge signal type and air pressure disturbance characteristics, the risk value of the insulated ring main unit is calculated:

[0155]

[0156] Among them, FX represents the risk value of the insulated ring main unit. Labels for the pseudo-color spectra of discharge signals collected in real time. This represents the predicted amplitude change over the J+1 time window.

[0157] It should be noted that the discharge signal category is numerically categorized. For example, for single-region discharges (e.g., bottom, middle, top, along the surface): the basic coding is defined as: bottom = 0.3, middle = 0.4, top = 0.5, along the surface = 0.7 (otherwise = 0). For cross-region discharges (e.g., bottom-middle, full area): the complexity weight is defined according to the number of areas covered: across 2 areas (e.g., bottom-middle) = 1.0, across 3 areas (full area) = 1.5; discharge intensity level: low intensity = 0.1, medium intensity = 0.2, high intensity = 0.3; discharge nature: stable discharge = 0.1, unstable discharge = 0.3. For example, when the label of the discharge signal pseudocolor spectrum is "medium intensity bottom stable discharge", then the label... =0.3+0.2+0.1=0.6.

[0158] The risk value of the insulated ring main unit is compared with a preset threshold. If the risk value exceeds the preset threshold, an early warning is issued, enabling fault diagnosis of the insulated ring main unit. This comparison is a crucial decision-making step in fault diagnosis. The risk value is a comprehensive quantitative indicator calculated by integrating discharge signal type and predicted values ​​of air pressure disturbance characteristics (including amplitude changes, slope, and response lag time). Its calculation formula integrates multi-dimensional information such as spatial distribution, energy characteristics, and temporal response, comprehensively reflecting the severity and development trend of insulation degradation within the equipment. The preset threshold is a risk threshold pre-set based on historical equipment operating data, a typical fault case library, and industry standards, used to classify the safety level of equipment operation (e.g., normal, warning, fault). When the calculated risk value of the insulated ring main unit exceeds the preset threshold, it indicates that the current discharge and air pressure disturbance characteristics of the equipment have exceeded the normal operating range, posing a high risk of insulation failure. At this time, the system triggers an early warning mechanism, issuing alerts to maintenance personnel through visual interface prompts and SMS notifications. Simultaneously, by combining the discharge signal type, the system accurately locates potential fault areas, providing clear guidance for subsequent maintenance decisions. This early warning mechanism based on multi-feature fusion and threshold discrimination can effectively overcome the limitations of single-parameter monitoring, realize early identification and hierarchical control of insulated ring main unit faults, and improve the intelligence and precision of power distribution network equipment operation and maintenance.

[0159] This paper proposes a fault diagnosis method for environmentally friendly gas-insulated ring main units. It collects partial discharge data and pressure disturbance data from the metal-enclosed gas chamber using multiple types of sensors, and uses an inverse distance weighting method to locate the discharge position. Discharge features such as frequency, repetition density, and phase angle, as well as pressure disturbance features, are extracted to construct a pseudo-color spectrum of the discharge signal. Convolutional neural networks (CNN) and long short-term memory networks (LSTM) are used to train and model the image features and temporal features, respectively. Finally, the risk value is calculated by fusing the discharge position category and the predicted pressure disturbance value, thus achieving fault early warning. This method improves the accuracy and timeliness of fault diagnosis for insulated ring main units through multi-physics data fusion and spatiotemporal feature analysis.

[0160] By combining the frequency center energy ratio and the high-frequency energy proportion to calculate the comprehensive frequency energy characteristic value, the energy distribution characteristics and frequency component proportion of the discharge signal can be effectively characterized. The frequency center energy ratio reflects the concentration of discharge energy near the center frequency, while the high-frequency energy proportion reflects the contribution of high-frequency energy to the overall energy. The fusion of the two can sensitively capture the differentiated impact of discharge type (such as corona discharge, surface discharge, etc.) on frequency characteristics, providing key parameters for the subsequent construction of a pseudo-color spectrum of the discharge signal containing energy characteristics, enhancing the discriminative power of the feature space, and thus improving the accuracy of the convolutional neural network model in identifying discharge types.

[0161] Generating a pseudo-color spectrum of the discharge signal by mapping the discharge signal location sequence, frequency-energy integrated feature value, repetition rate density, and phase angle features into a two-dimensional matrix is ​​a key step in transforming multi-dimensional abstract features into a visualized image. By mapping the spatial coordinates of the gas chamber to two-dimensional planar coordinates, determining the spatial location by the discharge location, and using the frequency-energy integrated feature value and repetition rate density as color channel parameters, a multi-dimensional feature map containing spatial distribution, energy intensity, time frequency, and phase characteristics can be constructed. This mapping method not only achieves spatial visualization of discharge features but also integrates multi-physical quantity information through color encoding, enabling convolutional neural networks to automatically extract texture, shape, and color distribution features from the spectrum. This effectively overcomes the shortcomings of traditional feature vector analysis in its insufficient utilization of spatial correlation information, providing richer input dimensions for discharge pattern recognition.

[0162] By combining discharge signal categories and predicted values ​​of gas pressure disturbance characteristics to calculate the risk value of insulated ring main units, a multi-dimensional quantitative assessment of equipment operating status is achieved. Discharge location categories reflect the spatial distribution characteristics of discharges, while predicted values ​​of gas pressure disturbance characteristics reflect the response characteristics of gas states to discharges. The fusion of these two factors encompasses both the spatial migration patterns of discharge development and the temporal information of changes in gas insulation performance. By establishing a risk value calculation model, the dynamic coupling relationship between spatial and temporal characteristics can be transformed into specific numerical values, which, when compared with preset thresholds, enable graded early warning of faults. This comprehensive assessment method overcomes the limitations of monitoring single physical quantities, providing a more comprehensive reflection of the insulation degradation process, offering maintenance personnel a scientific basis for decision-making, and improving the precision of equipment condition management.

[0163] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0164] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit, characterized in that: Includes the following steps: Step S1: Collect partial discharge data from the metal-sealed gas chamber to obtain a discharge signal dataset; Step S2: Extract features from the discharge signal dataset to obtain discharge signal feature data, which includes: time-frequency complex matrix, repetition rate density, and phase angle features; Step S3: Perform frequency energy center analysis and high-frequency energy analysis on the time-frequency complex matrix respectively to obtain the frequency center energy ratio and high-frequency energy proportion. Step S4: Combine the frequency center energy ratio and the high-frequency energy ratio to calculate the comprehensive frequency energy characteristic value; based on the discharge signal dataset, obtain the discharge signal position sequence; perform two-dimensional matrix mapping and annotation on the discharge signal position sequence, comprehensive frequency energy characteristic value, repetition rate density and phase angle characteristics to obtain the annotated discharge signal pseudo-color spectrum. Step S5: Input the labeled pseudo-color spectrum of the discharge signal into the convolutional neural network model to obtain the trained convolutional neural network model; collect the pressure disturbance data of the metal sealed air chamber, extract features, obtain the pressure disturbance feature data, and input it into the long short-term memory network model to obtain the trained long short-term memory network model. Step S6: Collect the pseudo-color spectrum of the discharge signal and the characteristic data of the air pressure disturbance in real time, and input them into the trained convolutional neural network model and the trained long short-term memory network model respectively. The corresponding output discharge signal category and air pressure disturbance characteristic prediction value are combined to calculate the risk value of the insulated ring main unit. Based on the risk value of the insulated ring main unit, the fault diagnosis of the insulated ring main unit is realized.

2. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 1, characterized in that: Feature extraction is performed on the discharge signal dataset to obtain discharge signal feature data, including the following specific steps: Frequency conversion is performed using short-time Fourier transform to obtain the time-frequency complex matrix of the discharge signal: ； in, Let L be the short-time Fourier transform output value of the m-th overlapping frame signal at frequency point k, and L be the length of each overlapping frame, 64. <L 4096, Let k be the m-th overlapping frame signal of the discharge signal, where m is the index of the m-th overlapping frame and k is the frequency index. K is the window function, K is the frame shift, and n is the index of the sampling point within the overlapping frame. Calculate the repetition density for each time window: ； in, For time repetition rate density, This represents the total number of discharge pulses within the time window. Let j be the total length of the j-th time window. = T represents the total acquisition time, and J represents the total number of time windows; Calculate the phase dispersion for each time window: ； in, For phase dispersion, This represents the total number of discharge pulses within the time window. , For the first The power frequency phase corresponding to the time of each discharge pulse occurrence , This is the index of the discharge pulse.

3. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 2, characterized in that: Frequency energy center analysis is performed on the time-frequency complex matrix to obtain the frequency center energy ratio, including the following steps: By performing frequency energy center analysis on the time-frequency complex matrix within each time window, the frequency center energy ratio is obtained: ； in, Here, M represents the frequency center energy ratio, M is the total number of overlapping frames, and m is the index of the m-th overlapping frame. Let B be the center frequency index of the m-th overlapping frame, B be the center bandwidth, L be the length of each overlapping frame, and k be the frequency index. This represents the short-time Fourier transform of the m-th overlapping frame signal at frequency point k.

4. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 3, characterized in that: The proportion of high-frequency energy is: ； in, Here, M represents the proportion of high-frequency energy, m is the total number of overlapping frames, m is the index of the m-th overlapping frame, L is the length of each overlapping frame, and k is the frequency index. This represents the short-time Fourier transform of the m-th overlapping frame signal at frequency point k. This is the frequency index corresponding to the high-frequency threshold.

5. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 4, characterized in that: The calculation of the comprehensive frequency energy characteristic value by combining the frequency center energy ratio and the high-frequency energy ratio includes the following steps: By combining the frequency center energy ratio and high-frequency energy proportion within each time window, the comprehensive frequency energy characteristic value is calculated: ； in, This is the comprehensive characteristic value of frequency and energy. The frequency center energy ratio, This represents the proportion of high-frequency energy.

6. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 5, characterized in that: The process of obtaining the discharge signal location sequence based on the discharge signal dataset includes the following specific steps: The discharge signal location sequence was calculated by performing location analysis on the discharge signal dataset within each time window using the inverse distance weighting method. ； in, Indicates within the time window Intrinsic coordinates ( The discharge probability under ( ) is given, where N is the total number of sensors, and N>

3. It is the first The position coordinates of each sensor It is the first Each sensor in the time window The maximum amplitude of the received discharge signal, It is a metal-enclosed air chamber space area. It is an indicator function, indicating that only if the point... The value is 1 when the chamber is enclosed in a metal cavity, and 0 otherwise. The discharge probability is compared with a preset discharge probability threshold. If it is greater than the preset threshold, it is marked as the cylindrical coordinate of the discharge signal. The cylindrical coordinates of all the filtered discharge signals are combined to obtain the discharge signal position sequence within each time window. The discharge signal position sequence of the nth time window is then obtained. The cylindrical coordinates of the discharge signals are .

7. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 6, characterized in that: The process of mapping and labeling the discharge signal position sequence, frequency-energy comprehensive feature value, repetition rate density, and phase angle feature into a two-dimensional matrix to obtain the labeled pseudo-color spectrum of the discharge signal includes the following specific steps: Map all points on the cylindrical coordinate system of the sealed metal chamber to two-dimensional coordinates to construct a pseudo-color spectrum of the discharge signal: ； in, For the first The two-dimensional horizontal axis of the pseudo-color spectrum of the discharge signal. For the first The azimuth angle of the cylindrical coordinate system of the discharge signal. The resolution of the two-dimensional matrix of the pseudo-color spectrum of the discharge signal; ； in, For the first The two-dimensional vertical axis of the pseudo-color spectrum of the discharge signal. For the first The height of the discharge signal cylindrical coordinate system, where H is the total height of the metal-sealed gas chamber. The resolution of the two-dimensional matrix of the pseudo-color spectrum of the discharge signal; In the cylindrical coordinate system of each discharge signal The ratio of the total radius R of the metal-enclosed gas chamber to the discharge signal pseudo-color spectrum is used as the first value in the spectrum. Two-dimensional coordinates of a discharge signal ( , Mapping of the red channel: ； in, For the first The mapping of the discharge signal in the red channel, For the first The radius of the discharge signal in the cylindrical coordinate system, R is the total radius of the metal-enclosed gas chamber, and clip() is the clip function; The frequency-energy comprehensive characteristic value of each discharge signal and the total number of times the discharge position sequence appears in all time windows within the total acquisition time T are used as the first value in the pseudocolor spectrum of the discharge signal. Two-dimensional coordinates of a discharge signal ( , Mapping of the green channel: ； in, For the first The mapping of a discharge signal in the green channel of the discharge signal pseudo-color spectrum. For the first The frequency-energy comprehensive characteristic value of each discharge signal, where clip() is the clip function. For the first The total number of times a discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T. This represents the maximum number of times the discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T; The phase dispersion, repetition density, and total number of times the discharge position sequence appears in all time windows within the total acquisition time T for each discharge signal are used as the index of the discharge signal pseudocolor spectrum. Two-dimensional coordinates of a discharge signal ( , Mapping of the green channel: ； in, For the first The mapping of a discharge signal in the blue channel of the discharge signal pseudocolor spectrum. For the first Phase dispersion of each discharge signal, For the first The repetition rate density of each discharge signal For the first The maximum number of pulses of a discharge signal within a time window, where clip() is the clip function. For the first The total number of times a discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T. This represents the maximum number of times the discharge signal appears in the discharge location sequence across all time windows of the total acquisition time T; The pseudo-color spectrum of the discharge signal is annotated using the discharge signal position sequence, and finally an annotated pseudo-color spectrum of the discharge signal is obtained.

8. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 7, characterized in that: The real-time collection of pseudo-color spectra of discharge signals and pressure disturbance feature data, and the input of these data into a trained convolutional neural network model and a trained long short-term memory network model, respectively, to output the discharge signal category and predicted values ​​of pressure disturbance features, includes the following specific steps: The pseudo-color spectrum of the discharge signal is collected in real time within a total time window of T, input into a trained convolutional neural network model, and the discharge signal category is output. ； in, For the prediction of the convolutional neural network model The probability of each label category. For the convolutional neural network model to the first The original predicted values ​​for each label are obtained from the output of the fully connected layer; The label with the highest predicted probability among all labels in the neural network model is selected as the label for the pseudo-color spectrum of the discharge signal collected in real time. , indicating the first One tag; The system collects and normalizes a sequence of air pressure disturbance feature data over a total time window of T, then inputs it into a trained long short-term memory network model to output predicted air pressure disturbance features. ； in, This represents the predicted values ​​of amplitude change, steepness slope, and response lag time over J+1 time windows. This is the weight matrix of the fully connected layer in a Long Short-Term Memory (LSTM) network model. This represents the hidden states of the Long Short-Term Memory network model at time T. This is the bias term for the fully connected layer in a Long Short-Term Memory (LSTM) network model.

9. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 8, characterized in that: The risk value of the insulated ring main unit is: ； Among them, FX represents the risk value of the insulated ring main unit. Labels for the pseudo-color spectra of discharge signals collected in real time. This represents the predicted amplitude change over the J+1 time window.

10. The method for diagnosing operational faults in an environmentally friendly gas-insulated ring main unit according to claim 9, characterized in that: The method of diagnosing faults in insulated ring main units based on their risk values ​​includes the following specific steps: The risk value of the insulated ring main unit is compared with a preset threshold. When the risk value of the insulated ring main unit is greater than the preset threshold, it indicates that the current discharge characteristics and air pressure disturbance characteristics have exceeded the normal operating range, and there is a high risk of insulation failure. This triggers the early warning mechanism, which provides a visual interface prompt and issues an alarm to the operation and maintenance personnel. At the same time, it combines the discharge signal type to accurately locate the potential fault area and provide clear guidance for subsequent maintenance decisions.

Images