GIS equipment gas leakage monitoring and early warning method based on confidence coefficient

Abstract

The invention belongs to the technical field of power equipment state monitoring, and discloses a GIS equipment gas leakage monitoring and early warning method based on confidence, which comprises the following steps: collecting and preprocessing gas monitoring data, and inhibiting fluctuation interference of original data; extracting peak value and valley value characteristics of the preprocessed data by adopting an envelope line method, and establishing a leakage trend identification model based on probability distribution; a tolerance mechanism and a confidence coefficient threshold are introduced, and effective leakage signals are screened; and outputting a leakage alarm result and a performance index. Smooth data peak and valley values are extracted through an envelope line method, the confidence coefficient of a'continuous peak and valley decrease 'mode is quantified in combination with a probability distribution model, a 0.001 MPa tolerance mechanism is introduced to avoid critical misjudgment, slow leakage with the daily average variation smaller than or equal to 0.001 MPa can be recognized, the detection rate is increased by 40% or above compared with a traditional threshold value method and a traditional slope method, and the requirement that the abnormal defect detection rate is larger than 80% is met.

Description

Technical Field

[0001] This invention belongs to the field of power equipment condition monitoring technology, specifically a confidence-based method for monitoring and early warning of gas leaks in GIS equipment. Background Technology

[0002] In power systems, SF6 gas, with its colorless, odorless, non-flammable, and inert chemical properties, as well as its excellent insulation and arc-extinguishing capabilities, is widely used in high-voltage electrical equipment such as GIS (Gas Insulation System) as an insulating and arc-extinguishing medium. However, SF6 gas decomposes under high-voltage arcs or high temperatures, producing highly toxic decomposition products such as SF4 and SOF2. Even trace leaks pose safety hazards. Moreover, SF6 gas is approximately five times denser than air. When leaks occur in indoor GIS equipment, SF6 gas and its decomposition products easily accumulate in lower levels, creating a locally oxygen-deficient and toxic environment, seriously threatening the lives of maintenance personnel.

[0003] SF6 gas leaks can be categorized into slow leaks and acute leaks, with significant differences between the two. Slow leaks are characterized by gradual concentration changes, long duration, and insidious initial signs, making them prone to missed detection by traditional monitoring methods due to data noise or algorithmic flaws. Acute leaks, on the other hand, are caused by sudden seal failure leading to a rapid increase in concentration, making them more likely to trigger threshold alarms. The difficulty of issuing warnings for these two types of leaks differs significantly. Therefore, accurately identifying and promptly issuing warnings about slow SF6 gas leaks is crucial.

[0004] In the prior art, patent application CN119290271B discloses an intelligent online monitoring system and method for SF6 gas leakage. This technical solution obtains characteristic values ​​by analyzing basic information, then acquires the internal safety factor of the equipment based on these characteristic values, and simultaneously analyzes the target gas concentration to obtain the external safety factor. Next, it analyzes the internal and external safety factors to determine the detection interval; finally, it issues operating commands based on the target gas concentration, thereby executing alarm operations and fan control. However, this solution neglects the characteristics of slow changes in the internal safety factor, small data decrease, and unclear data trend when SF6 leaks slowly from the combined electrical appliance. Furthermore, it does not consider situations with large data fluctuations. In such cases, it is difficult to achieve effective alarms through threshold comparison or incremental (slope) comparison methods.

[0005] Currently, SF6 gas density monitoring mainly relies on two types of devices: mechanical density relays and digital meters, combined with leak alarm algorithm design and setpoint parameter debugging. Equipment condition diagnosis requires testing and adjusting the algorithm's setpoint parameters based on a historical database of normal and abnormal defect states to achieve timely feedback on leak status. However, due to interference from the "low dynamic signal" of slow leaks, the distinguishability between normal and abnormal states in historical data is weakened, and the setpoint parameter debugging lacks accurate basis, leading to the risk of missed or false diagnoses. This not only directly threatens personal safety but also makes it difficult to guarantee the reliability of equipment insulation and arc-extinguishing performance. Summary of the Invention

[0006] The purpose of this invention is to provide a confidence-based method for monitoring and early warning of gas leaks in GIS equipment, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a confidence-based method for monitoring and early warning of gas leaks in GIS equipment, comprising the following steps: S1, Collection Gas monitoring data is preprocessed to suppress fluctuations and interference in the raw data. S2. The envelope method is used to extract the peak and valley features of the preprocessed data, and a leakage trend identification model based on probability distribution is established. S3. Introduce a tolerance mechanism and confidence threshold to filter effective leakage signals; S4. Output leakage alarm results and performance indicators.

[0008] Preferably, step S1 includes the following steps: A1. Collect SF6 gas pressure and temperature data using mechanical density relays or digital meters configured in GIS equipment, with a sampling period of 3-5 minutes to ensure the validity of peak and valley data; The pressure of SF6 gas is significantly affected by temperature. As the temperature rises, the gas expands and the pressure increases, and vice versa.

[0009] If only pressure data is used to determine a leak, misjudgment may occur due to temperature fluctuations (such as a sudden drop in ambient temperature causing a natural decrease in pressure, rather than a leak).

[0010] This algorithm synchronously collects temperature data and uses a temperature compensation algorithm (such as ideal gas law correction) to correct the pressure data, eliminating temperature interference and ensuring the accuracy of leak detection.

[0011] The specific steps are as follows: Data collection phase: Pressure values ​​are recorded synchronously every 3-5 minutes. and temperature value T; Temperature compensation calculation: Based on the reference temperature , calculate the compensated pressure value = × ; Analysis: Replace the original pressure data with the compensated pressure value .

[0012] A2. Smooth the original data based on the moving average method. The calculation formula is as follows:

[0013] Where, is the smoothed data value, is the j-th original data within the sliding window, and n is the window step size, which can be adjusted according to the fluctuation characteristics of the site data;

[0014] A3. Suppress the random fluctuations of the data through preprocessing, retain the trend characteristics of slow leakage, and output the denoised time series data.

[0015] Preferably, the S2 includes the following steps: B1. Based on the smoothed time series data , use the envelope method to extract the daily peak and valley values; B2. Construct the upper envelope: Interpolate the local maximum points of to form the upper boundary curve of the daily air pressure fluctuation; B3. Construct the lower envelope: Interpolate the local minimum points of[[ID=​​​​​​​​​​​​​​​Derivation of confidence level: Confidence = 1 - P, forming confidence threshold ranges for different modes; C3. Output the peak sequence {max1,max2,...,max7} and valley sequence {min1,min2,...,min7} within 7 days, along with the probability and confidence data of the corresponding decline pattern, to provide feature input for subsequent alarm judgment; Preferably, in step S3, if a more slow leakage problem is considered, the data for a fixed number of days can be merged.

[0017] Preferably, in step S3, in order to improve the reliability of alarms and reduce misjudgments caused by the peak and valley values ​​being close for two consecutive days, the algorithm also introduces the concept of tolerance, that is, it is considered that a downward trend can only be judged when the peak or valley value on the second day is significantly lower than the peak or valley value on the first day.

[0018] Preferably, when the tolerance is set to 0.001 MPa, a valid decrease is counted only when max1-max2>0.001 MPa.

[0019] Preferably, personalized configuration parameters are set according to the differences in data characteristics of different converter stations. The personalized parameters include moving average step size, confidence threshold and tolerance.

[0020] A confidence-based GIS equipment gas leak monitoring and early warning system includes: The data acquisition module is used to collect SF6 gas pressure and temperature data, with a sampling period of 3-5 minutes. The data preprocessing module is used to smooth the collected raw data based on the moving average method, suppress random fluctuations in the data, and output denoised time series data. The feature extraction module is used to extract the daily peak and valley features of the preprocessed data using the envelope method, construct the upper and lower envelopes, and extract the daily peak value maxk and valley value mink. The trend identification module is used to establish a leakage trend identification model based on probability distribution based on the extracted peak and trough features, define "effective downward trend", and calculate the trend confidence based on the probability distribution model; The alarm output module is used to introduce a tolerance mechanism and a confidence threshold to filter valid leakage signals and output leakage alarm results and performance indicators. The personalized configuration module is used to configure setpoint parameters in a personalized manner to take into account the differences in data characteristics of different converter stations. The setpoint parameters include moving average step size, confidence threshold and tolerance.

[0021] The beneficial effects of this invention are as follows: 1. The present invention extracts the peak and valley values of smooth data through the envelope line method, quantifies the confidence level of the "continuous peak and valley decreasing simultaneously" mode in combination with the probability distribution model, and introduces a 0.001 MPa tolerance mechanism to avoid critical misjudgment. It can identify slow leaks with a daily average change of ≤ 0.001 MPa, and the detection rate is increased by more than 40% compared with the traditional threshold method and slope method, meeting the requirement of "abnormal defect detection rate > 80%".

[0022] 2. Processed by the moving average method, the signal fluctuation of the present invention is effectively suppressed, and the data change trend is clearer, laying a foundation for subsequent analysis.

[0023] 3. Aiming at the differences in data waveforms of different converter stations, the present invention configures parameters such as the moving average step size and confidence threshold individually. The accuracy rate of the algorithm at each site is stable above 90%, solving the problems of poor adaptability of unified parameters and high false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the incoming line pipe bus 1D1GB-14 of the 500 kV first large group AC filter in Converter Station No. 1 of the present invention; Figure 2 Schematic diagram of the C-phase of the grounding knife gas chamber of the No. 4 synchronous condenser 500427 in Converter Station No. 2 of the present invention; Figure 3 Schematic diagram of the A-phase of the 5634 AC filter in Converter Station No. 3 of the present invention; Figure 4 Schematic diagram of the B-phase gas chamber of the 5624 circuit breaker in the AC filter yard of Converter Station No. 3 of the present invention; Figure 5 Schematic diagram of the No. 3 gas chamber of the lead wire of the No. 12 synchronous condenser at 500 kV in Converter Station No. 1 of the present invention; Figure 6 Schematic diagram of the C-phase of the 5621 circuit breaker in Converter Station No. 2 of the present invention; Figure 7 Schematic diagram of the B-phase of the pressure gauge of the D19GB-8 gas chamber in Converter Station No. 2 of the present invention; Figure 8 Schematic diagram of the B-phase of the pressure gauge of the D19GB-8 gas chamber in Converter Station No. 1 of the present invention; Figure 9 Schematic diagram of the C-phase of the 5621 circuit breaker in Converter Station No. 2 of the present invention; Figure 10 Schematic diagram of the present invention when max1 > max2 and min1 < min2 (convergent oscillation); Figure 11 Schematic diagram of the present invention when max1 > max2 and min1 > min2 (downward trend); Figure 12 Schematic diagram of the present invention when max1 < max2 and min1 > min2 (diffusive oscillation); Figure 13 Schematic diagram of max1 < max2 and min1 < min2 (upward trend) for the present invention. Detailed implementation manners

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0026] Embodiment 1

[0027] The embodiment of the present invention provides a method for monitoring and warning gas leakage of GIS equipment based on confidence, including the following steps: S1. Acquisition Collect gas monitoring data and perform preprocessing to suppress the interference of the fluctuation of the original data; S2. Use the envelope method to extract the peak and valley features of the preprocessed data, and establish a leakage trend recognition model based on probability distribution; S3. Introduce a tolerance mechanism and a confidence threshold to screen effective leakage signals; S4. Output the leakage warning result and performance indicators.

[0028] Figures 1 to 9 In this case, the horizontal axis is the date and the vertical axis is the gas content.

[0029] The normal data of the SF6 gas pressure of the combined electrical apparatus (as shown in Figure 1 , 2 ) has a large fluctuation range (the maximum volatility can reach 0.015 MPa, that is, 3%), strong randomness, and the peak and valley values do not appear at a fixed time. The rapid gas leakage data (as shown in Figure 3 , 4 ) has strong suddenness and a fast decline rate. The slow gas leakage data (as shown in Figures 5-7 ) has a small decline amplitude and an unclear trend. Through the above algorithm steps, accurate processing can be achieved for different types of data, especially effectively identifying slow gas leakage. For example, for the slow gas leakage data of phase C of the 5621 circuit breaker in the 2nd converter station ( Figure 6 ), after preprocessing, peak and valley value extraction, and confidence screening, its leakage trend can be accurately captured and the warning result can be output.

[0030] Embodiment 2

[0031] The steps included in the above S1 are as follows: A1. Collect SF6 gas pressure and temperature data using mechanical density relays or digital meters configured in GIS equipment, with a sampling period of 3-5 minutes to ensure the validity of peak and valley data; The pressure of SF6 gas is significantly affected by temperature. As the temperature rises, the gas expands and the pressure increases, and vice versa.

[0032] If only pressure data is used to determine a leak, misjudgment may occur due to temperature fluctuations (such as a sudden drop in ambient temperature causing a natural decrease in pressure, rather than a leak).

[0033] This algorithm synchronously collects temperature data and uses a temperature compensation algorithm (such as ideal gas law correction) to correct the pressure data, eliminating temperature interference and ensuring the accuracy of leak detection.

[0034] The specific steps are as follows: Data collection phase: Pressure values ​​are recorded synchronously every 3-5 minutes. and temperature value T; Temperature compensation calculation: based on reference temperature Calculate the pressure value after compensation. = × ; Analysis: Based on the compensated pressure value Replace the original stress data.

[0035] A2. The original data is smoothed using the moving average method, and the calculation formula is as follows:

[0036] in, The smoothed data values, The j-th original data point is represented by n, which is the window step size and can be adjusted according to the fluctuation characteristics of the station data.

[0037] A3. By preprocessing, random fluctuations in the data are suppressed, the trend characteristics of slow leakage are preserved, and the denoised time series data is output.

[0038] Taking the data from the pressure gauge of phase B in chamber D19GB-8 of converter station No. 1 on December 31, 2024 as an example ( Figure 8 The raw data was collected at a sampling period of 3-5 minutes, and a moving average was performed with every 10 data points as the window step size (n=10), which effectively suppressed data fluctuations; taking the data of phase C of circuit breaker 5621 at converter station No. 2 from November 27 to December 27, 2024 as an example ( Figure 9), After collecting at a sampling period of 3 - 5 minutes, a moving average is performed with a window step of 12 hours (n = the number of sampling points within 12 hours). After preprocessing, the maximum volatility of the data drops from 3% to below 0.5%. The downward trend of slow leakage is more obvious, and the output denoised time series data can be used for subsequent peak and valley value extraction.

[0039] Example 3

[0040] The S2 includes the following steps: B1. Based on the smoothed time series data , the envelope method is used to extract the daily peak and valley values; B2. Construct the upper envelope: Interpolate the local maximum points of to form the upper boundary curve of the daily air pressure fluctuation; B3. Construct the lower envelope: Interpolate the local minimum points of to form the lower boundary curve of the daily air pressure fluctuation; B4. Extract the eigenvalue: Denote the maximum value of the daily upper envelope as the daily peak maxk, and the minimum value of the lower envelope as the daily valley mink, obtaining consecutive daily feature pairs.

[0041] For the time series data of phase C of the 5621 circuit breaker at Converter Station No. 2 after moving average preprocessing ( Figure 9 the 12 - hour moving average data in

[0042] Example 4

[0043] The S3 includes the following steps: C1. Define "effective downward trend": When the air pressure characteristics for two consecutive days satisfy maxk + 1 < maxk and mink + 1 < mink, and the difference exceeds the tolerance Δ, it is determined as one effective decline; C2. Calculate the trend confidence level based on the leakage trend recognition model of probability distribution: Assume that when the gas density is stable, the random probability of a single group having two consecutive days of effective decline is 0.25; Enumerate all the combination patterns of "continuous effective decline" within 7 days (a total of 14 kinds), and calculate the occurrence probability according to the formula P = C×0.25^m×0.75^(7 - m); where C is the combination number and m is the number of effective declines; Derive the confidence level: Confidence = 1 - P, and form the confidence level threshold interval under different patterns; C3. Output the peak sequence {max1, max2,..., max7}, valley sequence {min1, min2,..., min7} within 7 days, as well as the probability and confidence level data of the corresponding decline pattern, providing characteristic inputs for subsequent alarm judgment.

[0044] Taking the slow air leakage data of Chamber 3 of the Lead of Phase Adjusting Machine No. 12 at 500 kV of Converter Station No. 1 ( Figure 5 ), extract the peak sequence {max1, max2,..., max7} and valley sequence {min1, min2,..., min7} within 7 days. If the air pressure characteristics for two consecutive days satisfy max2 < max1, min2 < min1 and the difference exceeds the tolerance Δ, it is determined as an effective decline. Assuming that the random probability of a single group of continuous two-day effective decline when the gas is stable is 0.25, if there is a pattern of "only 1 continuous 3-day effective decline" within 7 days (a total of 5 combinations), calculate according to the formula P = 5×0.25²×0.7 The calculated occurrence probability is 9.89%. Derive the confidence level = 1 - 9.89% = 90.11%, and form the confidence level threshold interval under this pattern; output the peak sequence, valley sequence within 7 days, as well as the probability (9.89%) and confidence level (90.11%) data of this decline pattern, providing characteristic inputs for judging whether there is leakage in this air chamber.

[0045] Example 5

[0046] In S3 above, if considering the problem of even slower air leakage, the data for a fixed number of days can be merged.

[0047] Regarding the extremely slow air leakage data of Phase B of the pressure gauge of Chamber D19GB - 8 of Converter Station No. 2 ( Figure 7 ), the air leakage process of this data is longer and the daily average change is smaller. It is difficult to capture the leakage trend by only extracting the peak and valley values on a single day. At this time, the data for a fixed number of days is merged, and the peak and valley values are taken every 4 days for judgment. That is, the maximum value of the upper envelope line for the 1st - 4th days is used as the peak value of this merged period, and the minimum value of the lower envelope line is used as the valley value of this period. The same applies to the 5th - 8th days, and so on, expanding the applicable range of the algorithm for air leakage detection from 7 days to 28 days (4×7), and successfully identifying the extremely slow leakage process of this air chamber.

[0048] Example 6

[0049] In S3, to improve alarm reliability and reduce misjudgments caused by the peak and valley values ​​being close for two consecutive days, the algorithm also introduces the concept of tolerance, that is, it is considered that a downward trend can only be judged when the peak or valley value on the second day is significantly lower than the peak or valley value on the first day.

[0050] When analyzing the air pressure data of phase A of AC filter 5634 at converter station No. 3, it was found that the peak values ​​(e.g., max1=0.87MPa, max2=0.868MPa) and valley values ​​were close for two consecutive days. Directly judging these values ​​could easily lead to misjudgments of a downward trend due to minor fluctuations. After introducing the concept of tolerance, a downward trend is only judged when the peak value of the second day is significantly lower than that of the first day, or when the valley value of the second day is significantly lower than that of the first day (e.g., the difference exceeds the set tolerance). This avoids misjudgments caused by minor numerical drifts due to instrument accuracy errors and small fluctuations in ambient temperature, and improves the reliability of leak alarms at this station.

[0051] Example 7

[0052] When the tolerance is set to 0.001MPa, a valid decrease is counted only when max1-max2>0.001MPa.

[0053] Normal air pressure data for phase C of the grounding switch gas chamber of synchronous condenser 500427 at converter station No. 2 ( Figure 2 Analysis revealed that the peak values ​​for some consecutive days were relatively small, such as max1 = 0.542 MPa and max2 = 0.5415 MPa. In this case, max1 - max2 = 0.0005 MPa < 0.001 MPa, which does not meet the "effective decrease" condition and is therefore not counted as an effective decrease. Furthermore, the slow leakage data of phase C of circuit breaker 5621 at this station ( Figure 6 The peak difference over two consecutive days was max1=0.702MPa and max2=0.699MPa. Since max1-max2=0.003MPa>0.001MPa, the condition was met and it was counted as an effective decrease, which accurately distinguished between normal data fluctuations and the actual leakage trend.

[0054] Example 8

[0055] To address the differences in data characteristics among different converter stations, personalized configuration parameters are set, including moving average step size, confidence threshold, and tolerance.

[0056] The data characteristics of different converter stations differ significantly. For example, the maximum fluctuation rate of normal data at converter station 1 reaches 0.015 MPa (3%), while the fluctuation rate of normal data at converter station 2 is relatively low. Based on this, personalized configuration parameters were set: for converter station 1, a tolerance of 0.0025 MPa, a confidence threshold of 90%, and a moving average step size of 120 minutes were selected, resulting in a false alarm rate of 0% and a recall rate of 83% after debugging; for converter station 3, a tolerance of 0.0025 MPa, a confidence threshold of 97%, and a moving average step size of 120 minutes were selected, resulting in a false alarm rate of 0% and a recall rate of 80%; for converter station 2, a tolerance of 0.002 MPa, a confidence threshold of 92%, and a moving average step size of 120 minutes were selected, resulting in a false alarm rate of 5% and a recall rate of 100%. All stations met the requirements of "detection rate > 80% and accuracy > 90%", solving the problem of poor adaptability of unified parameters.

[0057] Peak-valley analysis

[0058] Assuming constant gas density and random fluctuations in air pressure data, the peak and trough changes in air pressure over two consecutive days should be evenly distributed according to the following four scenarios. The probability of a continuous downward trend (i.e., both peak and trough values ​​decreasing simultaneously) over two consecutive days with constant gas density is 0.25 (1 / 4), while the probability for the other scenarios is 0.75 (3 / 4). Figures 10-13 )

[0059] Therefore, the probabilities and confidence levels of various scenarios where the peak and trough of air pressure decrease simultaneously within a week can be calculated, as follows: Scenario 1: Within 7 days, there is exactly one instance where the peak and trough values ​​decrease simultaneously over two consecutive days. There are 6 situations Probability of occurrence: p = 6 × 0.251 × 0.755 = 35.6% Confidence level: 1 - 35.6% = 64.4% Scenario 2: Within 7 days, there are exactly two consecutive days where the peak and trough values ​​decrease simultaneously. There are 10 situations Probability of occurrence: p = 10 × 0.252 × 0.754 = 19.78% Confidence level: 1 - 19.78% = 80.22% Scenario 3: Within 7 days, there is exactly one instance of a simultaneous decrease in both peak and trough values ​​over 3 consecutive days. There are 5 situations Probability of occurrence: p = 5 × 0.252 × 0.754 = 9.89% Confidence level: 1 - 9.89% = 90.11% Scenario 4: Within 7 days, there is one instance of a simultaneous decrease in both peak and trough values ​​over two consecutive days, and another instance of a simultaneous decrease in both peak and trough values ​​over three consecutive days. There are 12 situations Probability of occurrence: p = 12 × 0.253 × 0.753 = 7.91% Confidence level: 1 - 7.91% = 92.09% Scenario 5: Within 7 days, there are 4 consecutive days with both peak and trough values ​​decreasing simultaneously. There are 4 situations Probability of occurrence: p = 4 × 0.253 × 0.753 = 2.64% Confidence level: 1 - 2.64% = 97.36% Scenario 6: Within 7 days, there are 3 consecutive days where the peak and trough values ​​decrease simultaneously. There are 4 situations Probability of occurrence: p = 4 × 0.253 × 0.753 = 2.64% Confidence level: 1 - 2.64% = 97.36% Scenario 7: Within 7 days, there is one instance of a simultaneous decrease in both peak and trough values ​​over two consecutive days, and another instance of a simultaneous decrease in both peak and trough values ​​over four consecutive days. There are 6 situations Probability of occurrence: p = 6 × 0.254 × 0.752 = 1.32% Confidence level: 1 - 1.32% = 98.68% Scenario 8: Within 7 days, there are two consecutive 3-day peaks and troughs that decrease simultaneously. There are 3 situations Probability of occurrence: p = 3 × 0.254 × 0.752 = 0.66% Confidence level: 1 - 0.66% = 99.34% Scenario 9: Five consecutive days within 7 days show a simultaneous decrease in both peak and trough values. There are 3 situations Probability of occurrence: p = 3 × 0.254 × 0.752 = 0.66% Confidence level: 1 - 0.66% = 99.34% Scenario 10: Within 7 days, there are two consecutive days where the peak and trough values ​​decrease simultaneously, and one consecutive three days where the peak and trough values ​​decrease simultaneously. There are 3 situations Probability of occurrence: p = 3 × 0.254 × 0.752 = 0.66% Confidence level: 1 - 0.66% = 99.34% Scenario 11: Within 7 days, there are 6 consecutive days with both peak and trough values ​​decreasing simultaneously. There are two situations Probability of occurrence: p = 2 × 0.255 × 0.751 = 0.15% Confidence level: 1 - 0.15% = 99.85% Scenario 12: Within 7 days, there is one instance of a simultaneous decrease in both peak and trough values ​​over 3 consecutive days, and another instance of a simultaneous decrease in both peak and trough values ​​over 4 consecutive days. There are two situations Probability of occurrence: p = 2 × 0.255 × 0.751 = 0.15% Confidence level: 1 - 0.15% = 99.85% Scenario 13: Within 7 days, there is one instance of a simultaneous decrease in both peak and trough values ​​over two consecutive days, and another instance of a simultaneous decrease in both peak and trough values ​​over five consecutive days. There are two situations Probability of occurrence: p = 2 × 0.255 × 0.751 = 0.15% Confidence level: 1 - 0.15% = 99.85% Scenario 14: Peak and trough values ​​decrease simultaneously for 7 consecutive days There is one situation Probability of occurrence: p = 1 × 0.256 = 0.02% Confidence level: 1 - 0.02% = 99.98% The algorithm adjusts the alarm sensitivity by setting a confidence level requirement, thereby filtering out some data that may indicate a leak due to random fluctuations within a 7-day period, and achieving reliable identification of slow leak defects. For example, when the algorithm's confidence level requirement is adjusted to 99.5%, only four situations can trigger an alarm; when the confidence level requirement is adjusted to 99.9%, only one situation can trigger an alarm, namely, an alarm is issued only when the peak and trough values ​​decrease simultaneously for seven consecutive days.

[0060] The test samples were selected using both sampling rate screening and defect pattern screening. Sampling rate filtering The algorithm generates alerts by comparing daily peak and trough values. Currently, the sampling cycle for converter stations has been standardized to 3-5 minutes, ensuring the acquisition of effective peak and trough data. Some historical data had a lower sampling frequency (sampling intervals exceeding 30 minutes), which prevented accurate peak and trough data acquisition when actual air pressure fluctuations were significant, causing the algorithm to malfunction. Therefore, during test sample selection, only data with higher sampling rates were retained.

[0061] Defect pattern screening

[0062] The test data samples, filtered by sampling rate, are divided into two categories: positive samples and negative samples. Positive samples are data that are manually judged to have caused a slow leak defect, while negative samples are data that are manually judged not to have caused a leak defect. It is important to note that historical data showing rapid leak defects cannot be used as positive samples for the algorithm.

[0063] Due to significant differences in data characteristics among different converter stations, the setpoint parameters (moving average step size, confidence requirement, tolerance) and test sample sets for the alarm algorithm at each station were set independently. Specifically, there were 6 positive samples, 2 positive samples, 5 positive samples, and 0 positive samples for converter stations 1, 2, 3, and 4, respectively, and 100 negative samples for each of the four converter stations. Each set of data corresponds to 1 to 3 months of data from one monitoring point (approximately 15,000 to 43,000 data points per set).

[0064] By adjusting the settings at each station, the algorithm's detection rate was increased to over 80%, and its accuracy to over 90%. Some debugging results are shown in Table 1. The gray area in the table represents the optimized settings and test results of the SF6 gas slow leakage alarm algorithm for each converter station.

[0065] Table 1. Setting and commissioning results of the four converter stations

[0066] A confidence-based GIS equipment gas leak monitoring and early warning system includes: The data acquisition module is used to collect SF6 gas pressure and temperature data, with a sampling period of 3-5 minutes. The data preprocessing module is used to smooth the collected raw data based on the moving average method, suppress random fluctuations in the data, and output denoised time series data. The feature extraction module is used to extract the daily peak and valley features of the preprocessed data using the envelope method, construct the upper and lower envelopes, and extract the daily peak value maxk and valley value mink. The trend identification module is used to establish a leakage trend identification model based on probability distribution based on the extracted peak and trough features, define "effective downward trend", and calculate the trend confidence based on the probability distribution model; The alarm output module is used to introduce a tolerance mechanism and a confidence threshold to filter valid leakage signals and output leakage alarm results and performance indicators. The personalized configuration module is used to configure setpoint parameters in a personalized manner to take into account the differences in data characteristics of different converter stations. The setpoint parameters include moving average step size, confidence threshold and tolerance.

[0067] It should be noted that, in this document, relational terms such as "first" and "second" are used only 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 process, method, article, or apparatus.

[0068] 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 confidence-based method for monitoring and early warning of gas leaks in GIS equipment, characterized in that, It includes the following steps: S1, Collection Gas monitoring data is preprocessed to suppress fluctuations and interference in the raw data. S2. Use the envelope method to extract the peak and valley features of the preprocessed data, and establish a leakage trend recognition model based on probability distribution; S3. Introduce a tolerance mechanism and a confidence threshold to screen out effective leakage signals; S4. Output the leakage alarm result and performance indicators.

2. The confidence-based gas leak monitoring and early warning method for GIS equipment according to claim 1, characterized in that: The S1 includes the following steps: A1. Collect the SF6 gas pressure and temperature data through the mechanical density relay or digital meter configured by the GIS device. Set the sampling period to 3 - 5 minutes to ensure the effectiveness of peak and valley data; A2. Smooth the original data based on the moving average method. The calculation formula is as follows: in, The smoothed data values, The j-th original data point is represented by n, which is the window step size and can be adjusted according to the fluctuation characteristics of the station data. A3. Suppress the random fluctuations of the data through preprocessing, retain the trend features of slow leakage, and output the denoised time series data.

3. The confidence-based gas leak monitoring and early warning method for GIS equipment according to claim 1, characterized in that: The S2 includes the following steps: B1. Based on smoothed time series data The daily peak and trough values ​​were extracted using the envelope method. B2. Constructing the upper envelope: For Local maxima are interpolated to form the upper boundary curve of daily air pressure fluctuations; B3. Construct the lower envelope: For Interpolation of local minimum points is performed to form the lower boundary curve of daily air pressure fluctuations; B4. Extract feature values: Denote the maximum value of the daily upper envelope as the daily peak maxk, and the minimum value of the daily lower envelope as the daily valley mink, and obtain consecutive daily feature pairs.

4. The confidence-based gas leak monitoring and early warning method for GIS equipment according to claim 3, characterized in that: The S3 includes the following steps: C1. Define an "effective downward trend": When the gas pressure characteristics of two consecutive days satisfy maxk + 1 < maxk and mink + 1 < mink, and the difference exceeds the tolerance Δ, it is determined as an effective decline; C2. Calculate the trend confidence based on the leakage trend recognition model of probability distribution: Assume that when the gas density is stable, the random probability of a single group of two consecutive days of effective decline is 0.25; Enumerate all "consecutive effective decline" combination patterns within 7 days, and calculate the occurrence probability according to the formula P = C×0.25^m×0.75^(7 - m); where C is the combination number and m is the number of effective declines; Derive the confidence: Confidence = 1 - P, and form the confidence threshold interval under different patterns; C3. Output the peak sequence {max1, max2,..., max7}, valley sequence {min1, min2,..., min7} within 7 days, and the probability and confidence data of the corresponding decline patterns, providing feature inputs for subsequent alarm judgment.

5. The confidence-based gas leak monitoring and early warning method for GIS equipment according to claim 1, characterized in that: In S3, if considering the problem of slower air leakage, the data of a fixed number of days can be merged.

6. The confidence-based gas leak monitoring and early warning method for GIS equipment according to claim 1, characterized in that: In S3, the algorithm also introduces the concept of tolerance, that is, it is considered that there is a significant decrease in the peak or valley value of the second day compared to the peak or valley value of the first day before it can be judged as a downward trend.

7. The confidence-based gas leak monitoring and early warning method for GIS equipment according to claim 6, characterized in that: When the tolerance is set to 0.001 MPa, only when max1 - max2 > 0.001 MPa, it is counted as an effective decline.

8. The confidence-based gas leak monitoring and early warning method for GIS equipment according to claim 1, characterized in that: For the data characteristic differences of different converter stations, configure the setting parameters personalizedly. The setting parameters include the moving average step size, confidence threshold, and tolerance.

9. A confidence-based gas leak monitoring and early warning system for GIS equipment, characterized in that, It includes: A data acquisition module, used to collect SF6 gas pressure and temperature data, and set the sampling period to 3 - 5 minutes; A data preprocessing module, used to smooth the collected original data based on the moving average method, suppress the random fluctuations of the data, and output the denoised time series data; A feature extraction module, used to extract the daily peak and valley features of the preprocessed data by the envelope method, construct the upper envelope and lower envelope, and extract the daily peak maxk and valley mink; The trend recognition module is used to establish a leakage trend recognition model based on probability distribution based on the extracted peak and trough features, define "effective downward trend", and calculate the trend confidence based on the probability distribution model; The alarm output module is used to introduce a tolerance mechanism and a confidence threshold to filter valid leakage signals and output leakage alarm results and performance indicators. The personalized configuration module is used to configure setpoint parameters in a personalized manner to take into account the differences in data characteristics of different converter stations. The setpoint parameters include moving average step size, confidence threshold and tolerance.

Images