Pipeline valve leakage temperature and humidity signal self-adaptive dynamic threshold monitoring method

Abstract

The invention belongs to the technical field of nuclear equipment pipeline valve leakage temperature and humidity signal analysis, and particularly relates to a pipeline valve leakage temperature and humidity signal self-adaptive dynamic threshold monitoring method. Collecting data and calculating a data reference according to leakage-free data; firstly, a data trend is judged through linear regression; then setting an initial threshold value, and dividing leakage states into no leakage, concerned leakage, suspected leakage and leakage occurrence through the data reference and the initial threshold value; finally, different leakage state probabilities are calculated; initializing the weights of the reference methods 1 and 2, and comprehensively analyzing the leakage state; and updating the threshold values in the reference method 1 and the reference method 2 and the weights of the two reference methods through the real label and the diagnosis label, and outputting a final leakage state. According to the method, the dynamic threshold value is adaptively extracted through the temperature and humidity data collected in real time to effectively represent and judge the state of the pipeline valve, and support is provided for nuclear equipment pipeline valve leakage monitoring.

Description

Technical Field

[0001] The present invention belongs to the technical field of analyzing temperature and humidity signals for leakage of nuclear equipment pipeline valves, and particularly relates to a method for adaptively dynamically threshold monitoring of temperature and humidity signals for pipeline valve leakage. Background Art

[0002] In recent years, there have been many leakage problems such as running, dripping, and leaking of nuclear equipment pipeline valves at home and abroad. The existing leakage detection methods for nuclear equipment pipeline valves are mainly divided into two categories: one is to rely on post - inspection of the pipeline valve status, which cannot detect the leakage source in time for repair, and is likely to cause the deterioration of leakage accidents and more serious economic losses; the other is to rely on parameters such as system pressure and water level to judge leakage, but small leaks are difficult to cause changes in system parameters; on the other hand, due to the complexity of pipeline valve types and working conditions, the empirical judgment of system parameters such as pressure or the fixed - threshold alarm method cannot be applied to all pipeline and valve status judgments, and is prone to false alarms.

[0003] Through sorting out relevant domestic and foreign standards and combining on - site test experience, the present invention proposes a method for adaptively dynamically threshold monitoring of temperature and humidity signals for pipeline valve leakage. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for adaptively dynamically threshold monitoring of temperature and humidity signals for pipeline valve leakage, which can effectively characterize and judge the pipeline valve status by adaptively extracting dynamic thresholds from real - time collected temperature and humidity data, and provide support for the leakage monitoring of nuclear equipment pipeline valves.

[0005] To achieve the above - mentioned purpose, the technical solution adopted by the present invention is as follows:

[0006] A method for adaptively dynamically threshold monitoring of temperature and humidity signals for pipeline valve leakage:

[0007] 1. Data acquisition, and calculating the data reference according to the non - leakage data;

[0008] 2. Leakage judgment criterion method 1: First, determine the data trend through linear regression; then set the initial threshold value, and divide the leakage state into non - leakage, leakage of concern, suspected leakage, and occurred leakage through the data reference and the initial threshold value; finally, calculate the probabilities of different leakage states;

[0009] 3. Leakage judgment criterion method 2: First, design a rectangular window and set the initial values of the rectangular window length and moving step size; then calculate the data mean, standard deviation, and normalized mean within the rectangular window; then divide the leakage state into non - leakage, leakage of concern, suspected leakage, and occurred leakage through the relationship between the mean, the number of times of standard deviation increase, and the normalized mean and the initialization threshold; finally, calculate the probabilities of different leakage states;

[0010] 4. Comprehensive method: Initialize the weights of baseline methods 1 and 2 and comprehensively analyze the leakage status;

[0011] 5. Threshold and weight optimization: Update the thresholds and weights of baseline methods 1 and 2 using the true labels and diagnostic labels, and output the final leakage status.

[0012] Step 1: Data collection and data benchmark selection calculation; (1) Data collection: Use optical fiber sensors to collect pipeline temperature and humidity signals to obtain a one-dimensional time series sequence of temperature and humidity X = [x1, x2, ...x i ,...,x n ], X represents the temperature and humidity time series, x i represents the temperature and humidity measurement value at the i-th time point in the sequence; (2) Benchmark analysis: the average value of the temperature and humidity data of the first two sampling frequency lengths before the initial working condition without leakage is used as the data benchmark x base .

[0013] Step 2: Pipeline valve leakage status analysis based on linear regression analysis and status classification; (3) Regression analysis: Use linear regression to fit the temperature and humidity signal X to determine whether the trend is increasing, and calculate the slope b1 through regression. The formula is as follows:

[0014] X=b1t+b2

[0015] Where: b1—regression coefficient, reflecting the rate of change of temperature and humidity signal X over time t. If b1>0, it means that the data shows an increasing trend, b1<0 means that the data shows a decreasing trend, and b2 represents the intercept;

[0016] According to the dimensionless data benchmark x base Set the status classification, the expression is as follows:

[0017]

[0018] Where: state 1—no leakage; state 2—concerned leakage; state 3—suspected leakage; state 4—leakage occurred; threshold1—threshold 1; threshold2—threshold 2, and threshold1 < threshold2;

[0019] (5) Calculate the probability of different states, the expression is as follows:

[0020]

[0021] Where: —Different state probabilities, when i=1 Represents the probability of no leakage state, when i=2 Indicates the probability of leakage state, when i=3 Indicates the probability of a suspected leakage state. When i = 4 Indicates the probability of a leakage-occurring state; — The data lengths of different states, Indicates the data length of the non-leakage state, Indicates the data length of the leakage-concerned state, Indicates the data length of the suspected leakage state, Indicates the data length of the leakage-occurring state; N num — The total data length.

[0022] The initial value of threshold1 is set to 5, and the initial value of threshold2 is set to 10.

[0023] Step 3: Analysis of the leakage state of pipeline valves based on a window function dynamic threshold; (6) Design a rectangular window to move on the time-domain signal. Its window length L is related to the sampling frequency f s and its domain is 0.1f s : 0.1f s : 0.4f s , and this domain is reset according to the actual situation; its moving step S = 1 / 4L; when the window slides on the sequence, the starting position of each window moves forward by a step S. Therefore, the starting index i k of the k-th window satisfies: i k = 1 + (k - 1)S, where: k — the k-th window;

[0024] Calculate the mean and standard deviation of the signal in the rectangular window and normalize the mean. For each sliding window, starting from the starting position i k = 1 + (k + 1)S, the data within the window is [x ik , x ik+1 ,..., x ik+L-1 , and the mean u k within the window is calculated by the following formula:

[0025]

[0026] where: ik — the starting index of the window, L — the window length;

[0027] The normalized result n k of the window mean is calculated as follows:

[0028]

[0029] where: u k — the normalized mean within the window, which is a dimensionless quantity; x base — the data reference, x max — the maximum value of the range of the temperature and humidity sensor;

[0030] The standard deviation σ in the k-th window of the time series k is calculated by the following formula:

[0031]

[0032] where: σ k — the standard deviation in the k-th window of the time series; ik — the starting index of the window; L — the window length; u k — the normalized mean within the window.

[0033] (8) State analysis: When the mean and standard deviation increase continuously 5 times, but threshold3 ≤ n in the 5th window k <threshold4, it is prompted that leakage needs attention; when the mean and standard deviation increase continuously 5 times and threshold4 ≤ n in the 5th window k <threshold5, it is prompted that leakage is suspected; when the mean and standard deviation increase continuously 5 times and u in the 5th window k ≥threshold5, an alarm is issued for leakage; otherwise, it is prompted that there is no leakage. At the same time, when the mean and standard deviation increase continuously 10 times, but threshold3 ≤ u in the 10th window k <threshold4, it is prompted that leakage needs attention; when the mean and standard deviation increase continuously 10 times and threshold4 ≤ n in the 10th window k <threshold5, it is prompted that leakage is suspected; when the mean and standard deviation increase continuously 10 times and n in the 10th window k ≥threshold5, an alarm is issued for leakage; otherwise, it is prompted that there is no leakage. The above content is expressed as: If the mean and standard deviation within the window meet the following conditions in 5 consecutive increases or 10 consecutive increases, they are classified into different types:

[0034]

[0035] where: State 1 — no leakage; State 2 — attention to leakage; State 3 — suspected leakage; State 4 — leakage occurred; threshold3 — threshold 3; threshold4 — threshold 4, threshold5 — threshold 5, and threshold3 < threshold4 < threshold5.

[0036] The initial value of threshold3 is set to 0.2, the initial value of threshold4 is set to 0.4, and the initial value of threshold5 is set to 0.6.

[0037] (10) Window length adjustment and probability calculation: If the final determination indicates no leakage, the reference data remains unchanged. Using the data within the 10th window as the starting value, subsequent data is calculated, and the above loop judgment is repeated. The probabilities of various leakage states are calculated through the following formula:

[0038]

[0039] Where: — Probability of occurrence of various leakage states; — Number of occurrences of various leakage states; N total — Total number of events for all leakage states,

[0040] (11) Comprehensive judgment: The final leakage state probability is calculated through the following formula:

[0041]

[0042] Where: — Probability of occurrence of various leakage states determined by the comprehensive judgment of the two methods; W in1 — Weight of Method 1, with its initial value set to 0.5; — Probability of occurrence of various leakage states determined by Method 1; W in2 — Weight of Method 2, with its initial value set to 0.5; — Probability of occurrence of various leakage states determined by Method 2.

[0043] Step 4: Weight optimization: Perform adaptive weight adjustment on the leakage states obtained by different methods to comprehensively analyze the pipeline state, and perform feedback adjustment on the thresholds of different methods; (12) Assume that during actual use, N alarms are made, that is, there are N samples, and the true labels are TrueLabels = {T1, T2,..., T N}, and the leakage state diagnosis labels of each method are P re = {p m1 , p m2 ,..., p mN}, where m represents the method number, and the accuracy rate A m of each method is calculated as follows:

[0044]

[0045] Where: A m — Accuracy rate of method m; N — Total number of samples; t i — True label of the i-th sample; p mi — Prediction of method m for the i-th sample; l(t i = p mi ) — Indicator function, when t i= p mi When the value is 1 at time , the prediction is correct; otherwise, the value is 0 and the prediction is incorrect.

[0046] By comparing and analyzing the true label and the diagnostic label each time, when the true label is consistent with the diagnostic label, that is, when the leakage state diagnosis is correct, the threshold remains unchanged; when the true label is inconsistent with the diagnostic label, that is, when the leakage state diagnosis is incorrect, it is divided into two cases: 1) The severity of the true label leakage is less than that of the diagnostic label. For example, when the true label is no leakage and the diagnostic label is leakage occurred, then in Method 1, the values of threshold1 and threshold2 gradually increase by 1 until the threshold value when the diagnostic label is consistent with the true label is the updated value; in Method 2, the values of threshold3, threshold4, and threshold5 gradually increase by 0.1 until the threshold value when the diagnostic label is consistent with the true label is the updated value; 2) The severity of the true label leakage is greater than that of the diagnostic label. For example, when the true label is leakage occurred and the diagnostic label is no leakage, then in Method 1, the values of threshold1 and threshold2 gradually decrease by 1 until the threshold value when the diagnostic label is consistent with the true label is the updated value; in Method 2, the values of threshold3, threshold4, and threshold5 gradually decrease by 0.1 until the threshold value when the diagnostic label is consistent with the true label is the updated value.

[0047] (13) Calculation of the normalized weight: Weight W m is the ratio of the accuracy of this method to the accuracies of all methods, and the calculation formula is as follows:

[0048]

[0049] In the formula: W m — The normalized weight of method m; A m — The accuracy of method m; A1 + A2 + … + A M —— The sum of the accuracies of all methods. This method provides 2 leakage state judgment methods, that is, the updated weights of the 2 methods are:

[0050]

[0051] The beneficial effects achieved by the present invention are:

[0052] (1) Two leakage state determination benchmark methods are constructed, combining the advantages of dimensional and dimensionless indicators, and comprehensively analyzing using a moving rectangular window and an adjustable window length, effectively reducing the misdiagnosis rate caused by random fluctuations, etc.;

[0053] (2) Compared with existing fixed-threshold method models, etc., the present invention combines the advantages of two benchmark methods and utilizes on-site maintenance data to adaptively optimize the threshold and weight parameters of the benchmark methods by comparing and analyzing diagnostic labels and true labels. This method adaptively adjusts the method model based on on-site data and maintenance results, facilitating its use and promotion on different devices;

[0054] (3) Using probability to divide the leakage state into states such as "no leakage", "leakage of concern", "suspected leakage", and "occurred leakage" can provide effective maintenance guidance for on-site personnel. Description of the Drawings

[0055] Figure 1 is the flowchart of the adaptive dynamic threshold monitoring method for pipeline valve leakage temperature and humidity signals;

[0056] Figure 2 is the schematic diagram of the state classification of Benchmark Method 1;

[0057] Figure 3 is the schematic diagram of the relationship between the rectangular window of Benchmark Method 2 and the moving position of real-time monitoring data;

[0058] Figure 4 is the schematic diagram of the dynamic mean curve and standard deviation curve of the monitoring data of Benchmark Method 2 within the rectangular window. Detailed Embodiment

[0059] The present invention will be described in detail below in conjunction with the drawings and specific embodiments.

[0060] An adaptive dynamic threshold monitoring method for pipeline valve leakage temperature and humidity signals is as follows:

[0061] (1) Data collection, and calculation of the data benchmark based on non-leakage data;

[0062] (2) Leakage judgment Benchmark Method 1: First, determine the data trend through linear regression; then set the initial threshold value, and divide the leakage state into "no leakage", "leakage of concern", "suspected leakage", and "occurred leakage" through the data benchmark and the initial threshold value; finally, calculate the probabilities of different leakage states;

[0063] (3) Leakage judgment Benchmark Method 2: First, design a rectangular window and set the initial values of the window length and moving step of the rectangular window; then, calculate the mean, standard deviation, and normalized mean of the data within the rectangular window; then, divide the leakage state into "no leakage", "leakage of concern", "suspected leakage", and "occurred leakage" through the number of times of growth of the mean, standard deviation, and the relationship between the normalized mean and the initialization threshold; finally, calculate the probabilities of different leakage states;

[0064] (4) Comprehensive method: Initialize the weights of Benchmark Method 1 and 2, and comprehensively analyze the leakage state;

[0065] (5) Threshold and weight optimization: Update the thresholds in the baseline method 1 in step (2) and the baseline method 2 in step (3), and the weights of the two baseline methods in step (4) through the true label and the diagnosis label, and output the final leakage state.

[0066] The key points and the points to be protected in the present invention are:

[0067] (1) The present invention proposes an implementation process of an adaptive dynamic threshold analysis and alarm method for one-dimensional time series of temperature and humidity collected by sensors such as optical fibers.

[0068] (2) The present invention proposes a pipeline valve leakage state analysis method based on linear regression analysis and state classification, and a pipeline valve leakage state analysis method based on a window function dynamic threshold. The former designs a dimensional index based on reference data, and the latter designs a dimensionless index based on the normalized mean, and divides the leakage state into states such as "concerned about leakage", "suspected leakage" and "leakage occurred" through probability. Because the data changes caused by random fluctuations are not easy to change the overall probability distribution of the data, the above two methods can effectively avoid false alarms caused by random small fluctuations of numerical values caused by on-site interference; on the other hand, the pipeline valve leakage state analysis method based on the window function dynamic threshold combines the leakage state probability through different window length rectangular windows and the number of growth times, further reducing the false alarm rate of the leakage state.

[0069] (3) Based on the pipeline valve leakage state analysis based on linear regression analysis and state classification and the baseline method of the pipeline valve leakage state analysis based on the window function dynamic threshold, the present invention proposes a pipeline valve leakage temperature and humidity signal adaptive dynamic threshold monitoring method. This method comprehensively utilizes the pipeline valve leakage state judgment probabilities of the two baseline methods, and iteratively updates the thresholds of the two baseline methods through weight optimization, so as to adaptively monitor the leakage states of different types of pipeline valves, and it is easy to promote the application of monitoring methods and strategies.

[0070] Aiming at the problem that fixed thresholds in the current monitoring process of nuclear equipment pipeline valves are prone to cause false alarms, the present invention proposes a pipeline valve leakage temperature and humidity signal adaptive dynamic threshold monitoring method for nuclear equipment.

[0071] As Figure 1 shown, this method constructs two baseline methods, and combines the two baseline methods to adaptively extract dynamic thresholds through real-time collected temperature and humidity data to effectively characterize and judge the pipeline valve state, providing support for the leakage monitoring of nuclear equipment pipeline valves. The specific implementation steps are as follows:

[0072] Step 1: Data collection and calculation of data reference selection;

[0073] (1) Data acquisition: Use sensors such as optical fibers to collect pipeline humidity / temperature signals, and obtain the one-dimensional time series sequence X of temperature and humidity, X = [x1, x2,... x i ,... x n (X represents the time series sequence of temperature and humidity, and x i represents the temperature and humidity measurement value at the i-th time point in the sequence);

[0074] (2) Baseline analysis: Use the average value of the temperature and humidity data with a length of 2 sampling frequencies before the initial working condition without leakage as the data baseline x base ;

[0075] Step 2: Analysis of the leakage state of the pipeline valve based on linear regression analysis and state classification;

[0076] (3) Regression analysis: Use linear regression to fit the temperature and humidity signal X to determine whether the trend is increasing. The formula for calculating the slope b1 through regression is as follows:

[0077] X = b1t + b2

[0078] In the formula: b1 - regression coefficient (slope), reflecting the change rate of the temperature and humidity signal X with time t. If b1 > 0, it means the data shows an increasing trend; if b1 < 0, it means the data shows a decreasing trend. b2 represents the intercept.

[0079] (4) As Figure 2 shown, set the state classification according to the dimensional data baseline x base , and the specific expression is as follows:

[0080]

[0081] In the formula: State 1 - no leakage; State 2 - leakage to be concerned about; State 3 - suspected leakage; State 4 - leakage occurred; threshold1 - threshold 1; threshold2 - threshold 2, and thr. In this method, the initial value of threshold1 is set to 5, and the initial value of threshold2 is set to 10. Its updated value is determined iteratively according to the comparison between the diagnostic label and the true label in step 4.

[0082] (5) Calculate the probabilities of different states, and the specific expression is as follows:

[0083]

[0084] In the formula: - probability of different states. When i = 1 represents the probability of the no-leakage state. When i = 2 represents the probability of the leakage-to-be-concerned-about state. When i = 3 Indicates the probability of a suspected leakage state. When i = 4 Indicates the probability of a leakage occurring state; ——The data lengths of different states, Indicates the data length of the non - leakage state, Indicates the data length of the leakage - concerned state, Indicates the data length of the suspected leakage state, Indicates the data length of the leakage occurring state; N num ——The total data length.

[0085] Step 3: Analysis of the leakage state of pipeline valves based on the window function dynamic threshold;

[0086] (6) As Figure 3 shown, design a rectangular window to move on the signal in the time domain. Its window length L is related to the sampling frequency f s and its domain is 0.1f s : 0.1f s : 0.4f s , and this domain can be reset according to the actual situation; its moving step S = 1 / 4L. When the window slides on the sequence, the starting position of each window moves forward by a step S. Therefore, the starting index i k of the k - th window satisfies:

[0087] i k = 1+(k - 1)S

[0088] In the formula: k——The k - th window.

[0089] (7) Calculate the mean and standard deviation of the signal in the rectangular window, and normalize the mean.

[0090] As Figure 4 shown, for each sliding window, starting from the starting position i k = 1+(k + 1)S, the data within the window is [x ik , x ik+1 ,..., x ik+L-1 . The mean u k within the window can be calculated by the following formula:

[0091]

[0092] In the formula: ik——The starting index of the window, L——The window length.

[0093] The normalized result n of the window mean k is calculated as follows:

[0094]

[0095] In the formula: uk —— Normalized mean within the window, a dimensionless quantity; x base —— Data benchmark, x max —— Maximum value of the temperature and humidity sensor range.

[0096] The standard deviation σ in the k-th window of the time series k can be calculated by the following formula:

[0097]

[0098] In the formula: σ k —— Standard deviation in the k-th window of the time series; ik —— Starting index of the window; L —— Window length; u k —— Normalized mean within the window.

[0099] (8) State analysis: When the mean and standard deviation increase continuously 5 times, but threshold3 ≤ n in the 5th window k < threshold4, it is prompted that "leakage needs attention"; when the mean and standard deviation increase continuously 5 times and threshold4 ≤ n in the 5th window k < threshold5, it is prompted that "suspected leakage"; when the mean and standard deviation increase continuously 5 times and u in the 5th window k ≥ threshold5, an alarm of "leakage occurred" is given; otherwise, it is prompted that "no leakage". At the same time, when the mean and standard deviation increase continuously 10 times, but threshold3 ≤ u in the 10th window k < threshold4, it is prompted that "leakage needs attention"; when the mean and standard deviation increase continuously 10 times and threshold4 ≤ n in the 10th window k < threshold5, it is prompted that "suspected leakage"; when the mean and standard deviation increase continuously 10 times and n in the 10th window k ≥ threshold5, an alarm of "leakage occurred" is given; otherwise, it is prompted that "no leakage". Among them, the initial value of threshold3 is 0.2, the initial value of threshold4 is 0.4, and the initial value of threshold5 is 0.6. The above content can be expressed as: If the mean and standard deviation within the window meet the following conditions in 5 or 10 consecutive increases, they are classified into different types:

[0100]

[0101] Where: State 1 - no leakage; State 2 - leakage concerned; State 3 - suspected leakage; State 4 - leakage occurred; threshold3 - threshold 3; threshold4 - threshold 4, threshold5 - threshold 5, and threshold3 < threshold4 < threshold5. In this method, the initial value of threshold3 is set to 0.2, the initial value of threshold4 is set to 0.4, and the initial value of threshold5 is set to 0.6. Its updated value is iteratively determined according to the comparison between the diagnostic label and the true label in step 4.

[0102] (10) Window length adjustment and probability calculation: Adjust the window length according to (6) and calculate according to the process of (7)-(9) to calculate the probabilities of "no leakage", "leakage concerned", "suspected leakage" and "leakage occurred", and give the final judgment. If the final judgment indicates "no leakage", the reference data remains unchanged, and the data in the 10th window is used as the starting value to calculate the subsequent data, and the above loop judgment is repeated. The probabilities of various leakage states can be calculated by the following formula:

[0103]

[0104] Where: ——Probability of occurrence of various leakage states; ——Number of occurrences of various leakage states; N total ——Total number of events of all leakage states,

[0105] (11) Comprehensive judgment: The above two methods are comprehensively analyzed and judged, and the final leakage state probability can be calculated by the following formula:

[0106]

[0107] Where: ——Probability of occurrence of various leakage states determined by the comprehensive judgment of the two methods; W in1 ——Weight of method 1, and its initial value can be set to 0.5; ——Probability of occurrence of various leakage states determined by method 1; W in2 ——Weight of method 2, and its initial value can be set to 0.5; ——Probability of occurrence of various leakage states determined by method 2.

[0108] Step 4: Weight optimization: Comprehensively analyze the pipeline state by the adaptive weight adjustment method for the leakage states obtained by different methods, and feedback and adjust the thresholds of different methods, so as to make the model more accurate.

[0109] (12) Suppose there are N alarms during actual use, that is, there are N samples, and the true labels are TrueLabels = [T1, T2,..., T N , and the leakage status diagnosis labels of each method are P re = {p m1 , p m2 ,..., p mN}, where m represents the method number. The accuracy rate A m of each method is calculated as follows:

[0110]

[0111] In the formula: A m —— The accuracy rate of method m; N —— The total number of samples; t i —— The true label of the i-th sample; p mi —— The prediction of method m for the i-th sample; l(t i = p mi ) —— The indicator function, which takes the value of 1 (correct prediction) when t i = p mi , otherwise it takes the value of 0 (wrong prediction).

[0112] By comparing and analyzing the true label and the diagnosis label each time, when the true label is consistent with the diagnosis label, that is, when the leakage status diagnosis is correct, the threshold value does not change; when the true label is inconsistent with the diagnosis label, that is, when the leakage status diagnosis is wrong, it is divided into two cases: 1) The severity of the true label leakage is less than the severity of the diagnosis label leakage. For example, when the true label is no leakage and the diagnosis label is leakage occurred, then in method 1, the values of threshold1 and threshold2 gradually increase by 1 until the threshold value when the diagnosis label is consistent with the true label is the updated value; in method 2, the values of threshold3, threshold4, and threshold5 gradually increase by 0.1 until the threshold value when the diagnosis label is consistent with the true label is the updated value. 2) The severity of the true label leakage is greater than the severity of the diagnosis label leakage. For example, when the true label is leakage occurred and the diagnosis label is no leakage, then in method 1, the values of threshold1 and threshold2 gradually decrease by 1 until the threshold value when the diagnosis label is consistent with the true label is the updated value; in method 2, the values of threshold3, threshold4, and threshold5 gradually decrease by 0.1 until the threshold value when the diagnosis label is consistent with the true label is the updated value.

[0113] (13) Calculation of the normalized weight: After obtaining the accuracy rate, a weight can be assigned to each method. The weight W mis the ratio of the accuracy of this method to the accuracy of all methods, and the calculation formula is as follows:

[0114]

[0115] In the formula: W m —— The normalized weight of method m; A m —— The accuracy of method m; A1 + A2 + … + A M —— The sum of the accuracies of all methods. This design provides 2 leakage state judgment methods, that is, the updated weights of the 2 methods are:

[0116]

Claims

1. An adaptive dynamic threshold monitoring method for temperature and humidity signals of pipeline valve leakage, characterized in that: I. Data collection, and calculation of data benchmark according to non-leakage data; II. Leakage judgment benchmark method 1: First, determine the data trend through linear regression; then set the initial threshold value, and divide the leakage state into non-leakage, leakage concern, suspected leakage, and leakage occurred according to the data benchmark and the initial threshold value; finally, calculate the probabilities of different leakage states; III. Leakage judgment benchmark method 2: First, design a rectangular window, and set the initial values of the rectangular window length and moving step size; then calculate the mean value, standard deviation, and normalized mean value of the data within the rectangular window; then divide the leakage state into non-leakage, leakage concern, suspected leakage, and leakage occurred according to the mean value, the number of times of standard deviation increase, and the relationship between the normalized mean value and the initialization threshold; finally, calculate the probabilities of different leakage states; IV. Comprehensive method: Initialize the weights of benchmark methods 1 and 2, and comprehensively analyze the leakage state; V. Threshold and weight optimization: Update the thresholds in benchmark methods 1 and 2 and the weights of the two benchmark methods through the true label and the diagnostic label, and output the final leakage state.

2. The method for adaptively dynamically threshold monitoring of temperature and humidity signals for pipeline valve leakage according to claim 1, characterized in that: Step 1: Data acquisition and calculation of data benchmark selection; (1) Data acquisition: Use fiber optic sensors to collect the temperature and humidity signals of the pipeline, and obtain the one-dimensional time series X of temperature and humidity = [x1, x2,... x i ,..., x n , X represents the time series of temperature and humidity, and x i represents the temperature and humidity measurement value at the i-th time point in the sequence; (2) Benchmark analysis: Use the average value of the temperature and humidity data with a length of 2 sampling frequencies before the initial working condition without leakage as the data benchmark x base .

3. The method for adaptively dynamically threshold monitoring of temperature and humidity signals for pipeline valve leakage according to claim 2, wherein: Step 2: Analysis of pipeline valve leakage state based on linear regression analysis and state classification; (3) Regression analysis: Use linear regression to fit the temperature and humidity signal X, and judge whether the trend increases. Calculate the slope b1 through regression, and the formula is as follows: X = b1t + b2 In the formula: b1 - regression coefficient, reflecting the change rate of the temperature and humidity signal X with time t. If b1 > 0, it means the data shows an increasing trend; if b1 < 0, it means the data shows a decreasing trend. b2 represents the intercept; Based on the dimensional data reference x base Set the status classification, and the expression is as follows: In the formula: State 1 - non-leakage; State 2 - leakage concern; State 3 - suspected leakage; State 4 - leakage occurred; threshold1 - threshold 1; threshold2 - threshold 2, and threshold1 < threshold2; (5) Calculate the probabilities of different states, and the expression is as follows: In the formula: — Probability of different states. When i = 1 represents the probability of no leakage state. When i = 2 represents the probability of concerned leakage state. When i = 3 represents the probability of suspected leakage state. When i = 4 represents the probability of leakage occurrence state; — Data length of different states, represents the data length of no leakage state, represents the data length of concerned leakage state, represents the data length of suspected leakage state, represents the data length of leakage occurrence state; N num — Total data length.

4. The method for adaptively dynamically threshold monitoring of pipeline valve leakage temperature and humidity signals according to claim 3, characterized in that: The initial value of threshold1 is set to 5, and the initial value of threshold2 is set to 10.

5. The method for adaptively dynamically threshold monitoring of pipeline valve leakage temperature and humidity signals according to claim 3, wherein: Step 3: Analysis of the leakage state of pipeline valves based on the dynamic threshold of the window function; (6) Design a rectangular window to move on the time-domain signal, and its window length L is related to the sampling frequency f s and its domain is 0.1f s : 0.1f s : 0.4f s , and this domain is reset according to the actual situation; its moving step size S = 1 / 4L; when the window slides on the sequence, the starting position of each window moves forward by a step size S. Therefore, the starting index i k of the k-th window satisfies: i k = 1 + (k - 1)S, where: k - the k-th window; Calculate the mean and standard deviation of the signal in the rectangular window, and normalize the mean. For each sliding window, starting from the starting position i k = 1+(k + 1)S, the data within the window is [x ik , x ik+1 ,..., x ik+L-1 , and the mean u k within the window is calculated by the following formula: In the formula: ik - starting index of the window, L - window length; Window mean normalization result n k The calculation is as follows: where: u k — normalized mean within the window, a dimensionless quantity; x base — data reference x max — Maximum value of the temperature and humidity sensor range; The standard deviation σ in the k-th window of the time series k is calculated by the following formula: Where: σ k — Standard deviation in the k-th window of the time series; ik — Starting index of the window; L — Window length; u k — Normalized mean within the window.

6. The method for adaptively dynamically threshold monitoring of pipeline valve leakage temperature and humidity signals according to claim 5, characterized in that: (8) Status analysis: When the mean and standard deviation continuously increase 5 times, but threshold3 ≤ n within the 5th window k <threshold4, it indicates that leakage needs attention; when the mean and standard deviation continuously increase 5 times and threshold4 ≤ n within the 5th window k <threshold5, it indicates suspected leakage; when the mean and standard deviation continuously increase 5 times and u within the 5th window k ≥threshold5, an alarm indicates that leakage has occurred; otherwise, it indicates no leakage. At the same time, when the mean and standard deviation continuously increase 10 times, but threshold3 ≤ u within the 10th window k <threshold4, it indicates that leakage needs attention; when the mean and standard deviation continuously increase 10 times and threshold4 ≤ n within the 10th window k ≤threshold5, it indicates suspected leakage; when the mean and standard deviation continuously increase 10 times and n within the 10th window k ≥threshold5, an alarm indicates that leakage has occurred; otherwise, it indicates no leakage. The above content is expressed as: If the mean and standard deviation within the window meet the following conditions in 5 consecutive or 10 consecutive increases, they are classified into different types: In the formula: State 1 - non-leakage; State 2 - leakage concern; State 3 - suspected leakage; State 4 - leakage occurred; threshold3 - threshold 3; threshold4 - threshold 4, threshold5 - threshold 5, and threshold3 < threshold4 < threshold5.

7. The method for adaptively dynamically threshold monitoring of pipeline valve leakage temperature and humidity signals according to claim 6, characterized in that: The initial value of threshold3 is set to 0.2, the initial value of threshold4 is set to 0.4, and the initial value of threshold5 is set to 0.

6.

8. The method for adaptively dynamically threshold monitoring of pipeline valve leakage temperature and humidity signals according to claim 6, characterized in that: (10) Window length adjustment and probability calculation: If the final judgment indicates non-leakage, the benchmark data remains unchanged. Use the data within the 10th window as the starting value to calculate the subsequent data, and repeat the above loop judgment. The probabilities of various leakage states are calculated through the following formula: Wherein: — Probability of occurrence of various leakage states; — Number of occurrences of various leakage states; N total — Total number of events of all leakage states, (11) Comprehensive judgment: The final leakage state probability is calculated through the following formula: Wherein: — The occurrence probability of various leakage states determined by the comprehensive judgment of the two methods; W in1 — The weight of Method 1, whose initial value is set to 0.5; — The occurrence probability of various leakage states determined by Method 1; W in2 — The weight of Method 2, whose initial value is set to 0.5; — The occurrence probability of various leakage states determined by Method 2.

9. The method for adaptively dynamically threshold monitoring of pipeline valve leakage temperature and humidity signals according to claim 8, characterized in that: Step 4: Weight Optimization: Adaptive weight adjustment is performed on the leakage states obtained by different methods to comprehensively analyze the pipeline state, and feedback adjustment is made to the thresholds of different methods; (12) Assume that N alarms are made during actual use, that is, there are N samples, and the true labels are TrueLabels = {T1, T2,..., T N}, and the leakage state diagnosis labels of each method are P re = {p m1 , p m2 ,..., p mN}, where m represents the method number, and the accuracy rate A m of each method is calculated as follows: Where: A m — The accuracy rate of method m; N — The total number of samples; t i — The true label of the i-th sample; p mi — The prediction of method m for the i-th sample; l(t i = p mi ) — Indicator function, which takes the value of 1 when the prediction is correct (i.e., t i = p mi ), and 0 otherwise (when the prediction is incorrect); Compare and analyze the true label and the diagnostic label each time. When the true label is consistent with the diagnostic label, that is, when the leakage state diagnosis is correct, the threshold remains unchanged; when the true label is inconsistent with the diagnostic label, that is, when the leakage state diagnosis is incorrect, it is divided into two cases: 1) The severity of the true label leakage is less than the severity of the diagnostic label leakage. For example, when the true label is no leakage and the diagnostic label is leakage occurred, then in Method 1, the values of threshold1 and threshold2 gradually increase by 1 until the threshold value when the diagnostic label is consistent with the true label is the updated value; in Method 2, the values of threshold3, threshold4, and threshold5 gradually increase by 0.1 until the threshold value when the diagnostic label is consistent with the true label is the updated value. 2) The severity of the true label leakage is greater than the severity of the diagnostic label leakage. For example, when the true label is leakage occurred and the diagnostic label is no leakage, then in Method 1, the values of threshold1 and threshold2 gradually decrease by 1 until the threshold value when the diagnostic label is consistent with the true label is the updated value; in Method 2, the values of threshold3, threshold4, and threshold5 gradually decrease by 0.1 until the threshold value when the diagnostic label is consistent with the true label is the updated value.

10. The method for adaptively dynamically threshold monitoring of pipeline valve leakage temperature and humidity signals according to claim 9, wherein: (13)Calculation of normalized weight: weight W m is the ratio of the accuracy of this method to the accuracy of all methods, and the calculation formula is as follows: Where: W m — The normalized weight of method m; A m — The accuracy rate of method m; A1 + A2 + … + A M —— The sum of the accuracy rates of all methods. This method provides two leakage state judgment methods, that is, the weights updated by the two methods are:

Images