Safety valve state abnormity monitoring method and system based on edge calculation

Abstract

The invention relates to the technical field of abnormity monitoring, in particular to a safety valve state abnormity monitoring method and system based on edge calculation. The method comprises the following steps: obtaining and carrying out layered denoising to obtain a clean acoustic signal; performing frequency domain energy value calculation on the clean acoustic signal to obtain energy density data, and then calculating energy difference to obtain energy change data; carrying out risk analysis and frequency domain characteristic decomposition based on the energy change data to obtain a leakage characteristic component, calculating energy deviation and carrying out threshold comparison to obtain a deviation anomaly sequence; tracking statistics is carried out according to the deviation anomaly sequence, the anomaly duration is obtained, interference elimination is carried out, and the refining anomaly duration is obtained; performing confidence calculation and threshold comparison based on the refining abnormal duration to obtain a priority alarm sequence; historical data of the safety valve are obtained, threshold value correction is carried out, threshold value comparison is carried out after a corrected alarm threshold value is obtained, and an instant alarm signal is obtained. According to the invention, the leakage signal of the safety valve is accurately identified and alarmed.

Description

Technical Field

[0001] This invention relates to the field of anomaly monitoring technology, and in particular to a method and system for monitoring the anomaly status of a safety valve based on edge computing. Background Technology

[0002] In the field of industrial safety, with the development of edge computing technology, how to effectively capture abnormal equipment signals and improve the real-time level of status monitoring has become a core issue that needs to be addressed in the development of current safety valve monitoring technology.

[0003] In one existing technology, safety valve monitoring strategies primarily rely on preset fixed acoustic thresholds, frequency ranges, or energy limits for judgment. By monitoring the real-time acoustic parameters of the safety valve, an alarm mechanism is triggered when a certain indicator exceeds the preset range, issuing a warning signal or initiating an emergency response procedure to ensure production safety. However, the noise environment in industrial sites is becoming increasingly complex, and fixed strategies are difficult to adapt to the actual characteristics of weak leakage signals under different operating conditions, leading to increased false alarm rates or a greater risk of missed alarms. Furthermore, traditional solutions often rely on centralized data processing methods, requiring the transmission of large amounts of raw data to remote locations for analysis. This not only increases the communication burden but may also lead to delayed warnings due to network latency.

[0004] In summary, existing technologies lack the ability to distinguish weak leakage signals and background noise from safety valves, resulting in a decrease in the accuracy of potential risk identification. Summary of the Invention

[0005] This invention provides a method and system for monitoring the abnormal status of safety valves based on edge computing. It can dynamically monitor the abnormal status of safety valves and issue timely alarms, solving the problem that the existing technology lacks the ability to distinguish weak leakage signals and background noise from safety valves, resulting in a decrease in recognition accuracy.

[0006] Firstly, to address the aforementioned technical problems, this invention provides a method for monitoring the abnormal status of a safety valve based on edge computing, comprising: The original acoustic signal is acquired, and layered denoising is performed on the original acoustic signal to obtain a clean acoustic signal; The frequency domain energy value of the clean acoustic signal is calculated to obtain energy density data, and the energy difference is calculated based on the energy density data to obtain energy change data; Risk analysis is performed based on the energy change data to obtain the leakage risk frequency. Then, frequency domain feature decomposition is performed based on the leakage risk frequency to obtain leakage feature components. The energy deviation is calculated based on the leakage characteristic components to obtain an energy deviation sequence. A threshold comparison is then performed based on the energy deviation sequence to obtain an abnormal deviation sequence. The abnormal sequence is tracked and statistically analyzed to obtain the duration of the abnormality. Interference is then eliminated based on the duration of the abnormality to obtain the duration of the refined abnormality. The confidence level is calculated based on the duration of the refining anomaly to obtain the cumulative confidence level. The threshold is then compared based on the cumulative confidence level to obtain the priority alarm sequence. Historical data of the safety valve is acquired, and a threshold correction is performed based on the historical data to obtain a corrected alarm threshold. The corrected alarm threshold is then compared with the priority alarm sequence to obtain an immediate alarm signal.

[0007] In one optional implementation, the step of acquiring the original acoustic signal and performing layered noise reduction on the original acoustic signal to obtain a clean acoustic signal includes: The original acoustic signal is acquired, and wavelet decomposition is performed on the original acoustic signal to obtain a set of signal coefficients; Soft thresholding is performed on the set of signal coefficients to obtain filtered signal coefficients. The filtered signal coefficients are then combined in reverse to obtain a clean acoustic signal.

[0008] In one optional implementation, the step of calculating the frequency domain energy value of the clean acoustic signal to obtain energy density data, and calculating the energy difference based on the energy density data to obtain energy change data, includes: Short-time Fourier transform is performed on the clean acoustic signal to obtain frequency domain amplitude data, and energy value is calculated based on the frequency domain amplitude data to obtain energy density data; Based on the energy density data, a spectrum is generated and logarithmic scaling is performed to obtain energy spectrum data; The energy difference between adjacent frames is calculated based on the energy spectrum data to obtain energy change data.

[0009] In one optional implementation, the step of performing risk analysis based on the energy change data to obtain the leakage risk frequency, and then performing frequency domain feature decomposition based on the leakage risk frequency to obtain leakage feature components, includes: The energy change data is compared with a preset change threshold and marked to obtain a risk time map. The leakage risk frequency is obtained by comparing the risk time map with a preset risk model. Frequency domain features are extracted based on the leakage risk frequency to obtain preliminary feature components. These preliminary feature components are then weighted and filtered to obtain leakage feature components.

[0010] In one optional implementation, the step of calculating the energy deviation based on the leakage characteristic components to obtain an energy deviation sequence, and performing a threshold comparison based on the energy deviation sequence to obtain a deviation anomaly sequence, includes: The leakage characteristic component is compared with the preset low-temperature ambient noise baseline to obtain the energy deviation value. The deviation statistics within the preset sliding time window are calculated based on the energy deviation value to obtain the deviation window data. The deviation window data is compared with a preset deviation threshold. If the deviation window data exceeds the preset deviation threshold, it is marked to obtain a deviation anomaly sequence.

[0011] In one optional implementation, the step of tracking and statistically analyzing the deviation anomaly sequence to obtain the anomaly duration, and then performing interference removal based on the anomaly duration to obtain the refined anomaly duration, includes: The adjacent windows are merged according to the deviation anomaly sequence to obtain the merged anomaly sequence. The duration of the anomaly is obtained by time tracking and statistics based on the merged anomaly sequence. The duration of the anomaly is compared with the preset interference duration. If the duration of the anomaly is shorter than the preset interference duration, it is determined to be a non-steady-state fluctuation and is removed to obtain the refined anomaly duration.

[0012] In one optional implementation, the step of calculating confidence based on the duration of the refining anomaly to obtain a cumulative confidence score, and then performing a threshold comparison based on the cumulative confidence score to obtain a priority alarm sequence, includes: A leakage change sequence is obtained by jointly constructing the leakage change sequence based on the refining anomaly duration and the leakage characteristic components. The confidence level is calculated based on the leakage change sequence to obtain a confidence level sequence. The cumulative confidence level data is obtained by integrating the confidence level sequence over time. The cumulative confidence data is compared with a preset confidence threshold. If the cumulative confidence data exceeds the preset confidence threshold, it is marked to obtain a priority alarm sequence.

[0013] In one optional implementation, the steps of acquiring historical safety valve data, performing threshold correction based on the historical safety valve data to obtain a corrected alarm threshold, and comparing the corrected alarm threshold with the priority alarm sequence to obtain an immediate alarm signal include: Obtain historical data of the safety valve, perform correction calculations based on the historical data of the safety valve to obtain a confidence correction factor, and perform threshold correction based on the confidence correction factor to obtain a corrected alarm threshold; The priority alarm sequence is compared with the corrected alarm threshold. If the deviation between the corrected alarm threshold and the priority alarm sequence is less than the preset alarm tolerance, an alarm message is generated to obtain an immediate alarm signal.

[0014] Secondly, the present invention provides a safety valve status anomaly monitoring system based on edge computing, comprising: The signal preprocessing module is used to acquire the original acoustic signal and perform layered noise reduction on the original acoustic signal to obtain a clean acoustic signal. The energy analysis module is used to calculate the frequency domain energy value of the clean acoustic signal to obtain energy density data, and calculate the energy difference based on the energy density data to obtain energy change data; The feature extraction module is used to perform risk analysis based on the energy change data, obtain the leakage risk frequency, and perform frequency domain feature decomposition based on the leakage risk frequency to obtain leakage feature components. The deviation detection module is used to calculate the energy deviation based on the leakage characteristic components, obtain the energy deviation sequence, and perform threshold comparison based on the energy deviation sequence to obtain the deviation anomaly sequence; The duration filtering module is used to track and statistically analyze the deviation anomaly sequence to obtain the duration of the anomaly, and to remove interference based on the duration of the anomaly to obtain the refined anomaly duration. The confidence assessment module is used to calculate the confidence level based on the duration of the refining anomaly, obtain the cumulative confidence level, and perform threshold comparison based on the cumulative confidence level to obtain the priority alarm sequence. The threshold correction module is used to acquire historical data of the safety valve, perform threshold correction based on the historical data of the safety valve to obtain a corrected alarm threshold, and compare the corrected alarm threshold with the priority alarm sequence to obtain an immediate alarm signal.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains a clean acoustic signal by acquiring the original acoustic signal and performing layered denoising, and calculates the frequency domain energy value and analyzes the energy difference of the clean acoustic signal. This enables a precise grasp of the correlation between energy change and frequency domain characteristics during the abnormal process of the safety valve, thereby effectively solving the problem that the existing technology lacks the ability to extract deep features of weak leakage signals, resulting in a decrease in the accuracy of abnormal identification.

[0016] (2) This invention achieves intelligent adaptive filtering of non-steady-state fluctuation interference by tracking and statistically analyzing the deviation anomaly sequence and eliminating interference based on the duration of the anomaly, and by using a duration filtering mechanism to obtain the refined anomaly duration. This effectively solves the problem that existing technologies use fixed threshold judgments to make it difficult to distinguish between real leakage and transient interference, leading to delayed early warning response or frequent false alarms.

[0017] (3) This invention calculates the cumulative confidence level based on the duration of the refining anomaly, and performs threshold correction and alarm sequence comparison based on the historical data of the safety valve, thereby realizing the closed-loop feedback optimization of the monitoring strategy. This effectively solves the problem that traditional monitoring and control lacks historical data correction and adaptive adjustment mechanisms, and cannot guarantee the accuracy of long-term monitoring. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the safety valve status anomaly monitoring method based on edge computing provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a safety valve status anomaly monitoring system based on edge computing provided in an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] Reference Figure 1 This invention provides a method for monitoring the abnormal status of a safety valve based on edge computing, comprising: S11, acquire the original acoustic signal, and perform layered noise reduction on the original acoustic signal to obtain a clean acoustic signal; S12, calculate the frequency domain energy value of the clean acoustic signal to obtain energy density data, and calculate the energy difference based on the energy density data to obtain energy change data; S13, perform risk analysis based on the energy change data to obtain the leakage risk frequency, and perform frequency domain feature decomposition based on the leakage risk frequency to obtain leakage feature components; S14, calculate the energy deviation based on the leakage characteristic components to obtain the energy deviation sequence, and perform threshold comparison based on the energy deviation sequence to obtain the deviation anomaly sequence; S15, Track and statistically analyze the deviation anomaly sequence to obtain the duration of the anomaly, and remove interference based on the duration of the anomaly to obtain the duration of the refined anomaly. S16, Calculate the confidence level based on the duration of the refining anomaly to obtain the cumulative confidence level, and compare the threshold based on the cumulative confidence level to obtain the priority alarm sequence; S17, acquire historical data of the safety valve, perform threshold correction based on the historical data of the safety valve to obtain a corrected alarm threshold, and compare the corrected alarm threshold with the priority alarm sequence to obtain an immediate alarm signal.

[0021] In step S11, the original acoustic signal is acquired, and layered denoising is performed on the original acoustic signal to obtain a clean acoustic signal, including: The original acoustic signal is acquired, and wavelet decomposition is performed on the original acoustic signal to obtain a set of signal coefficients; Soft thresholding is performed on the set of signal coefficients to obtain filtered signal coefficients. The filtered signal coefficients are then combined in reverse to obtain a clean acoustic signal.

[0022] First, the raw acoustic signal is acquired. This raw acoustic signal is collected using an acoustic sensor positioned near the safety valve, with a sampling frequency set to 48kHz to ensure that the acoustic signal characteristics generated when the safety valve leaks are captured. The acquired acoustic signal is represented as a discrete time-series data, denoted as x(n), where n represents the sampling point number, n = 0, 1, 2, ..., N-1, and N is the total number of sampling points.

[0023] The original acoustic signal is decomposed using wavelet decomposition to obtain a set of signal coefficients. The basis function for the wavelet decomposition is the db4 wavelet of the Daubechies wavelet family, and the decomposition order is set to 6. In the i-th decomposition, the signal is decomposed into two parts according to the frequency range: an approximation coefficient layer cAi and a detail coefficient layer cDi. The approximation coefficient layer represents the low-frequency trend of the signal, and the detail coefficient layer represents the high-frequency fluctuation details of the signal. The (i+1)-th decomposition further decomposes the approximation coefficient layer cAi obtained in the i-th decomposition to obtain cAi+1 and cDi+1, and so on.

[0024] For example, in a single raw acoustic signal acquisition, the sampling frequency is 48kHz. According to the Nyquist sampling theorem, this sampling frequency can effectively capture all acoustic signal components in the frequency range of 0Hz to 24kHz. After six decompositions, six detail coefficient layers and one approximation coefficient layer are obtained. cD1 corresponds to a frequency range of 12kHz-24kHz; cD2 corresponds to a frequency range of 6kHz-12kHz; cD3 corresponds to a frequency range of 3kHz-6kHz; cD4 corresponds to a frequency range of 1.5kHz-3kHz; cD5 corresponds to a frequency range of 0.75kHz-1.5kHz; cD6 corresponds to a frequency range of 0.375kHz-0.75kHz; and cA6 corresponds to a frequency range of 0-0.375kHz.

[0025] The signal coefficient set includes wavelet decomposition coefficients at each level, denoted as {cA6, cD6, cD5, cD4, cD3, cD2, cD1}. This set contains 7 sets of coefficient sequences, each with a different length (number of sampling points). Specifically, the length of the coefficients at the i-th level is approximately half the length of the original signal. iAfter obtaining the signal coefficient set, soft thresholding is performed on the coefficients of each layer. Soft thresholding is a nonlinear denoising method. Its basic principle is to set wavelet coefficients smaller than a threshold to zero or shrink them, thereby suppressing noise components while retaining useful signal features. Specifically, for the wavelet coefficients of the i-th layer, the soft threshold λi is first determined. The soft threshold λ can be calculated using a general threshold formula: Where cDi represents the detail coefficients of the i-th layer, median() represents the median function, and NL represents the length of the wavelet coefficients of that layer. A soft thresholding function is applied to the wavelet coefficients of each layer, defined as follows: Where cDi(k) represents the original wavelet coefficients, cDi'(k) represents the wavelet coefficients after soft thresholding, sign() represents the sign function, and max() represents the maximum value function. For example, in a set of signal coefficients, the soft threshold λ3 of cD3 is 0.3, and a certain coefficient value is 0.85. Applying the soft thresholding function to this value yields a new coefficient value of 0.85 - 0.3 = 0.55. However, for the point with a coefficient value of 0.15, since its absolute value is less than the threshold of 0.3, the new coefficient value after applying the soft thresholding function is 0.

[0026] Further, a clean acoustic signal is obtained by inverse combination based on the selected signal coefficients. Inverse combination is a wavelet reconstruction process, which reconstructs the time-domain acoustic signal by reconstructing each layer of wavelet coefficients after soft-thresholding. Inverse combination starts from the deepest layer and proceeds upwards layer by layer. First, the 6th-layer approximation coefficients cA6' and 6th-layer detail coefficients cD6' are subjected to inverse wavelet transform to reconstruct the 5th-layer approximation coefficients. Then, the reconstructed 5th-layer approximation coefficients and 5th-layer detail coefficients cD5' are subjected to inverse wavelet transform to reconstruct the 4th-layer approximation coefficients. This process continues layer by layer. Finally, the reconstructed 1st-layer approximation coefficients and 1st-layer detail coefficients cD1' are subjected to inverse wavelet transform to obtain the complete reconstructed signal, i.e., the clean acoustic signal.

[0027] In step S12, the frequency domain energy value of the clean acoustic signal is calculated to obtain energy density data. Based on the energy density data, the energy difference is calculated to obtain energy change data, including: Short-time Fourier transform is performed on the clean acoustic signal to obtain frequency domain amplitude data, and energy value is calculated based on the frequency domain amplitude data to obtain energy density data; Based on the energy density data, a spectrum is generated and logarithmic scaling is performed to obtain energy spectrum data; The energy difference between adjacent frames is calculated based on the energy spectrum data to obtain energy change data.

[0028] It should be noted that the window function used in the short-time Fourier transform is the Hanning window, with a frame length of 1024 sampling points, resulting in a frame length of 1024 × 1 / 48 = 21.3 milliseconds. The frame shift parameter is set to half the frame length, i.e., 512 sampling points. For the m-th frame, a signal segment within the frame is extracted and multiplied by the Hanning window function to obtain a windowed signal segment. A fast Fourier transform is then performed on the windowed signal segment to obtain the frequency domain information of the window. For the frequency domain information of the m-th frame, K frequency components can be obtained, where K is the number of points in the fast Fourier transform, defaulted to 1024 points. By performing the above processing sequentially on all time windows, a two-dimensional frequency domain amplitude data matrix, denoted as A(m, k), can be obtained, where m represents the time frame index, ranging from 0 to M-1, and M is the total number of frames; k represents the frequency index, ranging from 0 to K-1. Each element A(m, k) of the matrix represents the amplitude value of the m-th frame time window at the k-th frequency component.

[0029] Furthermore, in the frequency domain signal, the energy value of a certain frequency component is equal to the square of its amplitude value. For each element in the frequency domain amplitude data matrix A(m,k), its square value is calculated to obtain the energy density data matrix E(m,k). Each element E(m,k) of the energy density data matrix represents the energy density value at the k-th frequency component within the m-th frame time window. For example, at a certain moment, at the 500th frequency component in the 10th frame, the amplitude value A(10,500) is 0.8, then the energy density value E(10,500) at that position is 0.8. 2 =0.64.

[0030] It is worth noting that the logarithmic scaling compression compresses data with a large dynamic range into a smaller range while preserving the relative relationships between the data. Specifically, a base-10 logarithmic transformation is performed on each element of the energy density data matrix to obtain the energy spectrum data matrix S(m, k). It should be noted that since the logarithmic function is undefined at zero, and the energy density data may contain values ​​close to or equal to zero, 0.001 needs to be added to all energy density values ​​before the logarithmic transformation to avoid mathematical errors when taking the logarithm. After logarithmic scaling compression, the energy spectrum data matrix S(m, k) is obtained.

[0031] It should be noted that for the m-th and m+1-th frames, the difference between the energy spectrum data of the two frames is calculated for each frequency component k to obtain the energy change value. The energy change data can be represented as a matrix D(m, k), where D(m, k) = S(m+1, k) - S(m, k). For example, when the safety valve is working normally, the energy spectrum data of the 200th frequency component in three consecutive frames are 3.5, 3.6, and 3.5, respectively. The energy change values ​​between adjacent frames are 3.6-3.5=0.1 and 3.5-3.6=-0.1, respectively, with the change values ​​fluctuating around zero, indicating that the safety valve is stable at this time. However, when the safety valve begins to leak slightly, the energy spectrum data of the 200th frequency component is detected to be 3.5, 4.2, and 4.8 in three consecutive frames, with the energy change values ​​between adjacent frames being 4.2-3.5=0.7 and 4.8-4.2=0.6, respectively, which significantly exceeds the normal operating conditions.

[0032] In step S13, risk analysis is performed based on the energy change data to obtain the leakage risk frequency. Frequency domain feature decomposition is then performed based on the leakage risk frequency to obtain leakage feature components, including: The energy change data is compared with a preset change threshold and marked to obtain a risk time map. The leakage risk frequency is obtained by comparing the risk time map with a preset risk model. Frequency domain features are extracted based on the leakage risk frequency to obtain preliminary feature components. These preliminary feature components are then weighted and filtered to obtain leakage feature components.

[0033] It should be noted that the preset change threshold is a judgment benchmark value pre-set based on the energy fluctuation characteristics of the safety valve under normal operating conditions, used to distinguish between normal energy fluctuations and abnormal energy changes. The default setting is 0.5. Each element in the energy change data matrix D(m,k) is compared with the preset change threshold one by one. When the absolute value of the energy change value at a certain position is greater than the preset change threshold, that position is marked and recorded as a risk point. After marking, all risk points are organized to obtain a risk time map.

[0034] The risk time map is a two-dimensional label matrix R(m, k), where the matrix elements take values ​​of 0 or 1. When R(m, k) equals 1, it indicates that an abnormal energy change was detected at the k-th frequency component within the m-th frame time window; when R(m, k) equals 0, it indicates that the energy change at that location is within the normal range. The risk time map allows for a direct observation of the distribution characteristics of abnormal energy changes in both time and frequency dimensions.

[0035] It is worth noting that the aforementioned preset risk model is a reference model established by artificially creating a small leak in the safety valve under experimental conditions, and by collecting multiple sets of actual leak data and performing spectral analysis. The experimental conditions refer to a low-temperature, high-pressure working environment where the ambient temperature is controlled within the range of -50℃±2℃ and the pressure is maintained at 10MPa±0.5MPa, which can reflect the typical performance of real leak events in the frequency domain. Data results show that the leak signal is mainly concentrated in a frequency range with a center frequency of 6kHz and a bandwidth of 1kHz, and the leak causes a significant bulge in the energy density distribution within this frequency range. Subsequently, the risk time spectrum is compared with the preset risk model. For each time frame m in the risk time spectrum, the number of frequency components marked as risk points within the leakage characteristic frequency range (5.5kHz to 6.5kHz) is counted. When the proportion of the number of risk points within the leakage characteristic frequency range to the total number of risk points in a certain time frame exceeds a preset proportion threshold, the frequency distribution corresponding to that time frame is determined to conform to the leakage characteristics and is recorded as the leakage risk frequency. The preset ratio threshold is set to 60%, meaning that a frame is considered to have a leakage risk only when more than 60% of the risk points in the frame are concentrated in the leakage characteristic frequency range.

[0036] For example, in the risk time map of frame 105, a total of 20 risk points were detected, of which 15 risk points were located in the frequency range of 5.5kHz to 6.5kHz, accounting for 75%, which exceeded the preset proportion threshold of 60%. Therefore, frame 105 was determined to have leakage risk, and its corresponding frequency distribution was recorded as the leakage risk frequency. In frame 150, although 18 risk points were also detected, only 8 of them were located in the leakage characteristic frequency range, accounting for only 44%, which did not reach the preset proportion threshold. Therefore, this frame was not determined to have leakage risk.

[0037] It should be noted that the frequency domain feature extraction refers to selectively extracting frequency components and their energy information related to the leakage risk frequency from the complete frequency domain data. Specifically, for each time frame determined to have a leakage risk, the energy values ​​of all frequency components within the leakage characteristic frequency range are extracted from the energy spectrum data of that frame, forming a preliminary feature component sequence. The preliminary feature component sequence can be represented as a vector F(k), where the value range of k corresponds to the frequency index within the leakage characteristic frequency range. Each element F(k) of the vector represents the energy spectrum data value of the k-th frequency component in the current time frame.

[0038] Further, the leakage feature components are obtained by weighted screening of the preliminary feature components. The local spectrum weighted screening is based on the frequency distribution characteristics of the leakage signal in a preset risk model, assigning different weight coefficients to different frequency positions to highlight the frequency components with the most significant leakage features. In the weighted screening process, a spectrum weighting function W(f) is defined to describe the weight coefficient corresponding to each frequency f. The spectrum weighting function adopts a modified Gaussian distribution function: Among them, W min The minimum weight for the edge region is set to 0.8; W max The maximum weight for the center frequency is set to 1.5; f represents the current frequency value in kHz; f c The center frequency of the Gaussian distribution, i.e., the frequency corresponding to the peak position of the function, is set to 6.0kHz; σ represents the standard deviation of the Gaussian distribution, used to control the width of the distribution curve, and is set to 0.35kHz. In actual calculation, since the initial characteristic component sequence uses k as a variable, the spectral weighting function W(f) needs to be converted into W(k) first. For example, the leakage characteristic frequency range is 5.5kHz to 6.5kHz. The frequency index corresponding to 5.5kHz is calculated to be 5.5k÷24k×1024=235; the frequency index corresponding to 6kHz is 6k÷24k×1024=256. Then in W(k), we have W(235)=0.8+0.7×exp[-(0.5)² / (2×0.35²)]≈0.8+0.25≈1.05, W(256)=0.8+(1.5-0.8)×exp[0]=1.5. The leaked feature component F'(k) is obtained by performing element-wise multiplication of the initial feature component sequence F(k) with the spectral weighting function W(k).

[0039] In step S14, the energy deviation is calculated based on the leakage characteristic components to obtain an energy deviation sequence. A threshold comparison is then performed based on the energy deviation sequence to obtain an abnormal deviation sequence, including: The leakage characteristic component is compared with the preset low-temperature ambient noise baseline to obtain the energy deviation value. The deviation statistics within the preset sliding time window are calculated based on the energy deviation value to obtain the deviation window data. The deviation window data is compared with a preset deviation threshold. If the deviation window data exceeds the preset deviation threshold, it is marked to obtain a deviation anomaly sequence.

[0040] The preset low-temperature environmental noise baseline is a reference benchmark established under experimental conditions ensuring no leakage of the safety valve. The experimental conditions are the same as described in step S13. Acoustic signal data is continuously collected for one hour under the experimental conditions. The collected long-term acoustic data is processed according to the aforementioned steps S11 and S12, sequentially undergoing wavelet decomposition and reconstruction, short-time Fourier transform, energy density calculation, and logarithmic scaling compression to finally obtain the experimental condition energy spectrum. Statistical analysis is performed on the experimental condition energy spectrum in the time dimension, calculating the average energy of each frequency component over the one-hour time period to obtain the noise baseline vector B(k). Here, k represents the frequency index, with values ​​corresponding to all frequency components within the leakage characteristic frequency range. Each element B(k) of the vector represents the average energy level of the k-th frequency component under normal operating conditions.

[0041] Subsequently, the energy deviation of the leakage characteristic component F'(m,k) at the current monitoring moment is calculated. The unit of energy value is dB. The energy deviation value ΔE(m,k) is defined as the difference between the leakage characteristic component at the current moment and the noise baseline at the corresponding frequency. When the energy deviation value is positive, it indicates that the energy level of the current frequency component is higher than the normal operating baseline, and there may be an abnormal leakage; when the energy deviation value is negative or close to zero, it indicates that the current energy level is within the normal range.

[0042] For example, in frame 180 of the monitoring process, the frequency component with frequency index k=245 has a leakage characteristic component value F'(180, 245)=4.8 after local spectrum weighting filtering, while the noise baseline value B(245)=3.2. Therefore, the calculated energy deviation value ΔE(180, 245)=4.8-3.2=1.6, which is higher than normal operating conditions, indicating a potential leakage risk. Within the same frame, the frequency component with frequency index k=245 has F'(180, 250)=3.1, while the noise baseline value B(250)=3.3. The energy deviation value ΔE(180, 250)=3.1-3.3=﹣0.2, indicating that the energy level of this frequency component is slightly lower than the normal baseline, falling within the normal fluctuation range.

[0043] Furthermore, the purpose of introducing a sliding time window mechanism is to smooth the energy deviation value over time, avoiding misjudgments caused by instantaneous fluctuations at a single time point, while also capturing the persistent characteristics of the leakage signal. The preset sliding time window length is set to L frames, with a default value of 10. For the current time frame m, a sliding window centered on this frame is defined, with the window range from the (mL / 2)th frame to the (m+L / 2)th frame. Within this window range, the energy deviation value of each frequency component k is statistically analyzed, and the root mean square (RMS) of all deviation values ​​of that frequency component within the window is calculated as the deviation statistical value of that frequency component in the current window. The deviation window data is represented as a matrix RMS(m,k), where each element of the matrix represents the statistical intensity of the energy deviation of the kth frequency component at the m-th frame within the sliding window.

[0044] Further, the deviation window data is compared with a preset deviation threshold. If the deviation window data exceeds the preset deviation threshold, it is marked to obtain a deviation anomaly sequence. The preset deviation threshold θ(k) is a discrimination criterion determined based on statistical analysis of a large amount of experimental data, used to distinguish between normal energy fluctuations and abnormal leakage signals. Since energy fluctuations under experimental conditions usually follow an approximately normal distribution, the mean and standard deviation under experimental conditions are calculated for each frequency component, and the preset deviation threshold is set as the sum of the mean and three times the standard deviation of the position of that frequency component. In the actual monitoring process, for each time frame m and each frequency component k, the current deviation window data RMS(m,k) is compared with the corresponding preset deviation threshold θ(k) one by one. When RMS(m,k)>θ(k), it is determined that there is a deviation anomaly at that time and frequency point, and the deviation anomaly matrix A is entered into the list. abnormal In (m, k), the corresponding position is marked as 1; when RMS(m, k) ≤ θ(k), the time frequency point is determined to be in a normal state and marked as 0. After marking, the deviation anomaly sequence is obtained.

[0045] In step S15, the abnormal sequence is tracked and statistically analyzed to obtain the duration of the abnormality. Based on this duration, interference is eliminated to obtain the refined abnormal duration, including: The adjacent windows are merged according to the deviation anomaly sequence to obtain the merged anomaly sequence. The duration of the anomaly is obtained by time tracking and statistics based on the merged anomaly sequence. The duration of the anomaly is compared with the preset interference duration. If the duration of the anomaly is shorter than the preset interference duration, it is determined to be a non-steady-state fluctuation and is removed to obtain the refined anomaly duration.

[0046] It should be noted that during the merging of adjacent windows, the anomalous sequences are first statistically processed in the frequency dimension. For each time frame m, the number of all frequency components marked as anomalous within that frame is counted, denoted as N. abnormal (m). Further, define the frame-level anomaly detection criteria. When the number of abnormal frequency components N within a frame at a certain time... abnormal (m) exceeds the preset frame anomaly threshold N threshold If the frame is abnormal, it is marked as 1; otherwise, it is marked as normal and 0. The preset frame abnormality threshold N... threshold The threshold value N is set based on the total number of frequency components K within the leakage characteristic frequency range, with a default value of 30% of K. For example, if the leakage characteristic frequency range is 5.5kHz to 6.5kHz, the sampling frequency is 48kHz, and the FFT points are 1024, then the number of frequency components K corresponding to this frequency range is 21. Therefore, the preset frame anomaly threshold N... threshold =21 × 0.3 ≈ 6. Through the above frame-level anomaly determination, a simplified one-dimensional anomaly marker sequence F is obtained. abnormal (m), where m represents the time frame index, and the sequence elements take values ​​of 0 or 1. Subsequently, an adjacent window merging operation is performed on the one-dimensional anomaly marker sequence. When the interval between two anomaly frames is less than a preset merging interval threshold, these two anomaly frames and all frames in between are merged into a continuous anomaly time period. The preset merging interval threshold T... merge The default setting is 5 frames.

[0047] Specifically, starting from the beginning of the sequence, when the first abnormal frame is detected, its time index is recorded as m. start Continue searching forward; if in m... start The T after that merge The next abnormal frame m was detected within the frame range. node Then m start to m node All frames between m are marked as abnormal, when m start After T merge A new abnormal frame m was detected within the frame range. nod When the merge scope is extended to m nod and in m node The next T is initiated at the transition starting point. merge Range search; if in T merge If no abnormal frame is detected within the frame range, the current abnormal segment is determined to have ended, and m is recorded. node为 End position m end This forms a complete abnormal time period [m] start m end ].

[0048] For example, in a certain segment of monitoring data, the one-dimensional anomaly marker sequence is: [0, 0, 1, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 0, 0, ...], and the corresponding time frame indices are [100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, ...]. The first anomaly is detected in the 102nd frame, and m is set... start =102. In m start Anomalies were detected in frames 105 and 106 within the subsequent 5-frame range (i.e., frames 103 to 107). Therefore, frames 102 to 106 were merged into a single continuous anomalous time period, and m was set. node =106. Continuing the search, an anomaly was detected in frame 108. Therefore, frames 102 to 108 were merged into a single continuous anomaly time period, and m was set. node =108. Continuing the search, no anomalies were detected in the range of frames 109 to 113, indicating the end of the anomaly segment. m end =108, resulting in an abnormal time period of [102, 108]. Through adjacent window merging operations, the merged abnormal sequence C is obtained. abnormal , represented as a set of multiple abnormal time periods: C abnormal ={m start1 m end1 ], [m start2 m end2 ],...,[m start3 m end3 ]}, where N represents the total number of abnormal time periods after merging.

[0049] It is worth noting that the time tracking statistics calculate the number of frames for each abnormal time period and convert it into the actual physical time length. For the i-th abnormal time period [m] in the merged abnormal sequence... start m end The number of frames L that lasts framei =m end -m start +1, the frame length was calculated to be 21.3 milliseconds in step S12, then the duration T of the i-th abnormal time period is... duration，i =21.3×L framei For example, in an abnormal time period from frame 252 to 258, the duration is L. frame1 =258-252+1=7 frames, corresponding to anomaly duration T_duration1=21.3×7=149.1 milliseconds.

[0050] It should be noted that the preset interference duration T interferenceThe discrimination criterion is determined based on statistical analysis of transient interference signal characteristics under actual industrial operating conditions, and is set to 150 milliseconds by default. During interference removal, the set of abnormal durations T is... duration Each element T in duration，i Perform a comparison one by one. If T duration，i <T interference If T is abnormal, the abnormal time period is determined to be a transient interference signal and removed from the set of abnormal durations; if T duration，i ≥T interference If the abnormal time period is identified as a potential leakage event, it will be retained. After interference removal, a refined set of abnormal durations T' is obtained. duration .

[0051] In step S16, confidence is calculated based on the duration of the refining anomaly to obtain a cumulative confidence score. A threshold comparison is then performed based on the cumulative confidence score to obtain a priority alarm sequence, including: A leakage change sequence is obtained by jointly constructing the leakage change sequence based on the refining anomaly duration and the leakage characteristic components. The confidence level is calculated based on the leakage change sequence to obtain a confidence level sequence. The cumulative confidence level data is obtained by integrating the confidence level sequence over time. The cumulative confidence data is compared with a preset confidence threshold. If the cumulative confidence data exceeds the preset confidence threshold, it is marked to obtain a priority alarm sequence.

[0052] First, a leakage change sequence is obtained by jointly constructing the leakage change sequence based on the refined anomaly duration and the leakage feature components. The joint construction extracts feature data of all frames within the i-th anomaly time period from the leakage feature component data F'(m,k). Specifically, for the time range [m... start m end For each frame m, extract the energy values ​​F'(m,k) of all frequency components k within the leakage characteristic frequency range of that frame, and calculate the comprehensive energy characteristic value E of that frame. feature，i (m), the formula for calculating the comprehensive energy characteristic value is as follows: Where, k min and k max These represent the start and end frequency indices of the leakage characteristic frequency range, namely 5.5kHz and 6.5kHz, respectively. W(k) represents the spectral weighting function defined in step S13. Further, a time-dimensional statistical analysis is performed on the comprehensive energy characteristic values ​​of all frames within the i-th abnormal time period to calculate the average energy characteristic value E for that time period. mean，i The formula for calculating the average energy characteristic value is as follows: Through the above processing, the information of the i-th abnormal time period is integrated into a triplet data structure: [T' duration，i E mean，i m start，i ], representing the duration of the anomaly, the average energy characteristic value, and the starting time frame index, respectively. All N triplet data points for anomaly time periods are sorted according to their starting time frame index m. start，i Arranged in ascending order, the leakage change sequence L is obtained. sequence .

[0053] Further, confidence levels are calculated based on the leakage change sequence to obtain a confidence level sequence. The confidence level calculation employs a weighted fusion method, comprehensively considering the contributions of both the anomaly duration and energy feature components, and integrating multi-dimensional information by assigning different weight coefficients. For the i-th anomaly time period in the leakage change sequence, its constituent indicators are first normalized. For the anomaly duration T'... duration，i The time-normalized value T is calculated using an exponential normalization function. norm，i The exponential normalization function is defined as follows: Among them, T ref For reference duration, the default setting is 500 milliseconds. Since the longer the anomaly duration, the greater the possibility of leakage, when the anomaly duration approaches 0, the normalized value is close to 0; when the anomaly duration increases, the normalized value gradually increases and approaches 1; when the anomaly duration reaches or exceeds 3 times the reference duration, the normalized value is close to saturation, above 0.95.

[0054] For the average energy characteristic value E mean，i The process employs a linear normalization method. First, based on the experimental condition energy spectrum from step S14, the energy characteristic values ​​for each frame within one hour are calculated and statistically analyzed to obtain experimental condition characteristic data. Based on this experimental condition characteristic data, the energy baseline value E is determined. baseline and energy saturation value E saturation The energy reference value E baseline The energy saturation value E is set as the median of the characteristic data for 1 hour of experimental conditions. saturation The 95th percentile of the characteristic data for one hour of experimental conditions is set. Energy normalized value E norm，i The calculation formula is: Here, min() is the minimum value function, and max() is the maximum value function. This formula ensures that the normalized value is within the interval [0, 1]. When the average energy characteristic value is lower than the energy benchmark value, the normalized value is 0; when the average energy characteristic value reaches or exceeds the energy saturation value, the normalized value is 1; when it is between the two, the normalized value increases linearly.

[0055] After normalizing all indicators, a weighted fusion method is used to calculate the confidence value C for the i-th abnormal time period. confidence，i To emphasize the importance of anomaly duration, a weighting coefficient α of anomaly duration is set to 0.6, and a weighting coefficient β of the energy feature component is set to 0.4. The confidence score is then calculated using the following formula: For example, in a confidence calculation, the duration T' of a certain abnormal time period. duration，1 The time taken is 319.5 milliseconds, and the normalized value of the calculation duration T is calculated. norm，1 =1-exp(﹣319.5 / 500)=1-0.528=0.472. The average energy characteristic value E for this time period. mean，1 =6.35, assuming an energy baseline value E baseline =4.5, energy saturation value E saturation =9.0, then the energy normalization value E norm，1 =(6.35-4.5) / (9.0-4.5)=1.85 / 4.5=0.411. Final confidence score C confidence，1 =0.6×0.472+0.4×0.411=0.448.

[0056] Furthermore, time integration is performed on the confidence sequence to obtain cumulative confidence data. This time integration process aims to capture the cumulative effect of leakage states; when multiple abnormal time periods occur consecutively within a short time interval, the cumulative confidence will increase rapidly, thereby triggering a higher-level early warning response. Time integration is implemented using a sliding cumulative window mechanism with a window length of m. window Set to 100 frames, which is 2130 milliseconds.

[0057] For the j-th anomalous time period in the confidence sequence, its cumulative confidence C cumulative，j The calculation needs to consider all data within the time window T that occurred before this time period. window The confidence contribution of abnormal time periods within the range. Specifically, determine the start time frame m of the j-th abnormal time period. start，j Calculate the starting boundary frame m of the time window. window =m start，j -100. Traverse the confidence sequence and select all starting time frames m. start，i ≥mwindow For the abnormal time periods where i ≤ j, sum their confidence values ​​to obtain the cumulative confidence C. cumulative，j .

[0058] For example, in a certain monitoring, the confidence scores for five abnormal time periods were obtained. The starting frames and confidence scores for each time period were: [200, 0.386], [350, 0.512], [520, 0.628], [650, 0.455], and [780, 0.591]. For the third abnormal time period (starting frame 520), the time window range is all time periods after 520-100=420 frames. The second time period (starting frame 350) and the third time period itself satisfy the condition, therefore the cumulative confidence score C is... cumulative，3 =0.512+0.628=1.140.

[0059] The cumulative confidence level data is obtained through time integration. Finally, the cumulative confidence level data is compared with a preset confidence threshold. If the cumulative confidence level data exceeds the preset threshold, it is marked, resulting in a priority alarm sequence. The preset confidence threshold C... threshold This threshold is a criterion determined based on the safety valve leakage risk level and the acceptable false alarm rate in the industrial setting, with a default setting of 1. The setting of this threshold follows these principles: when the cumulative confidence level is below 1, it is judged as a low-risk state or a possible false alarm, and no alarm is triggered; when the cumulative confidence level reaches or exceeds 1, it is judged as a suspected leakage event and marked as a priority alarm event. For each element C in the cumulative confidence level data sequence... cumulative，j Perform a comparison one by one, if C cumulative，j ≥C threshold Then, the j-th abnormal time period is marked as a priority alarm event, and its start time frame m is recorded. start，j Cumulative confidence level C cumulative，j And the corresponding leakage change sequence information. All abnormal time periods that meet the conditions are sorted from highest to lowest cumulative confidence level to obtain the priority alarm sequence P. alarm .

[0060] In step S17, historical data of the safety valve is acquired, threshold correction is performed based on the historical data to obtain a corrected alarm threshold, and an immediate alarm signal is obtained by comparing the corrected alarm threshold with the priority alarm sequence, including: Obtain historical data of the safety valve, perform correction calculations based on the historical data of the safety valve to obtain a confidence correction factor, and perform threshold correction based on the confidence correction factor to obtain a corrected alarm threshold; The priority alarm sequence is compared with the corrected alarm threshold. If the deviation between the corrected alarm threshold and the priority alarm sequence is less than the preset alarm tolerance, an alarm message is generated to obtain an immediate alarm signal.

[0061] Specifically, first, count the total number N alarm events within the historical time window. total The historical time window is set to the last 7 days by default, which is a time range of 7 × 24 × 3600 seconds backward from the current moment. Within this time window, all events marked as priority alarms are extracted based on the alarm history data, and their total number is counted.

[0062] Furthermore, the number N of historical alarm events that were manually confirmed as actual leaks was counted. true And the number N of events that were identified as false alarms. false Based on these statistics, the historical alarm accuracy rate P is calculated. accuracy =N true / N total In addition, missed reporting also needs to be considered. Missed reporting refers to situations where a leak event actually occurred, but the monitoring system failed to issue an alarm in a timely manner. The identification of missed reporting events is based on maintenance record data and leak records discovered during manual inspections. The total number N of actual leak events occurring within the historical time window is calculated. leak (Including those detected by the system and those not detected), calculate the false negative rate P. miss =(N leak -N true ) / N leak .

[0063] Based on the above statistical parameters, the confidence correction factor δ is calculated. This confidence correction factor is calculated using a piecewise linear function, based on 100 sets of historical data with a confidence level of 95%, and is defined as follows: When P accuracy ≥0.85 and P miss When the value is ≤0.1, the system is considered to be in good condition, and the confidence correction factor δ=1.0, indicating that there is no need to adjust the current threshold.

[0064] When P accuracy <0.85 and P miss If the value is greater than 0.1, it is determined that there is a malfunction in the alarm program, and the administrator or maintenance personnel need to be notified to maintain the alarm program. When P accuracy When the confidence level is less than 0.85, a false alarm risk is identified, and the alarm threshold needs to be increased to reduce false alarms. The formula for calculating the confidence correction factor at this point is: When P missWhen the confidence level is greater than 0.1, a risk of missed alarms is identified, and the alarm threshold needs to be lowered to improve sensitivity. The formula for calculating the confidence correction factor at this point is: It should be noted that, to avoid drastic fluctuations in the correction factor leading to system instability, the range of the correction factor is limited to [0.7, 1.3]. When the calculated correction factor exceeds this range, it is automatically truncated to the boundary value. For example, in a certain correction calculation, 50 priority alarm events (N) triggered in the last 7 days are counted. total =50), of which 42 were confirmed as actual leaks (N true =42), 8 were confirmed as false alarms (N) false =8). According to maintenance records, 45 leakage events actually occurred during this period (N leak =45), of which 3 were not detected by the system. Calculate the historical alarm accuracy P. accuracy =42 / 50=0.84, calculate the false negative rate P. miss =(45-42) / 45=0.067. Since P accuracy <0.85 but P miss If the value is ≤0.1, the confidence correction factor is calculated using the accuracy correction formula: δ=1.0+0.5×(0.85-0.84)=1.005.

[0065] Further, the threshold is corrected based on the confidence correction factor to obtain the corrected alarm threshold. First, the baseline alarm threshold C is determined. threshold_base The baseline alarm threshold is a standard judgment benchmark determined based on experimental data and expert experience during the initial system design phase, and is set to 1.0. The corrected alarm threshold C... threshold_adjusted =δ×C threshold_base Subsequently, each abnormal time period in the priority alarm sequence is compared one by one. For the j-th abnormal time period, its cumulative confidence C is extracted. cumulative，j , and the corrected alarm threshold C threshold_adjusted The difference is calculated to obtain the deviation value Δ. alarm，j =C cumulative，j -C threshold_adjusted Then, the calculated deviation value Δ alarm，j With respect to preset alarm tolerance ε tolerance A comparison is made. The preset alarm tolerance is a buffer zone set to avoid frequent misjudgments of critical states, with a default value of 0.05. When Δ alarm，j <﹣ε tolerance When the cumulative confidence level for the abnormal time period is lower than the corrected alarm threshold, the alarm condition is not met, and no alarm signal is triggered. When Δ alarm，j ≥εtolerance When the cumulative confidence level of the abnormal time period exceeds the modified alarm threshold, the alarm triggering condition is met, alarm information is generated, and an immediate alarm signal is obtained.

[0066] For example, in a certain comparison process, the priority alarm sequence contains three abnormal time periods with cumulative confidence levels of 0.95, 1.08, and 1.18, respectively. The current corrected alarm threshold C threshold_adjusted =1.005, preset alarm tolerance ε tolerance =0.05. For the first abnormal time period, calculate the deviation value Δ_alarm, 1=0.95-1.005=-0.055<-0.05, therefore the alarm condition is not met and no alarm is triggered. For the second abnormal time period, calculate the deviation value Δ alarm，2 =1.08-1.005=0.075≥0.05, therefore the alarm condition is met, and the alarm signal is triggered. For the third abnormal time period, calculate the deviation value Δ. alarm，3 =1.18-1.005=0.175≥0.05, the alarm condition is met, alarm information is generated, and an immediate alarm signal is obtained.

[0067] In summary, this invention provides a safety valve status anomaly monitoring method based on edge computing, which can dynamically monitor the abnormal status of the safety valve and issue timely alarms. This solves the problem that the existing technology lacks the ability to distinguish weak leakage signals and background noise from safety valves, resulting in a decrease in recognition accuracy.

[0068] Reference Figure 2 This invention provides a safety valve status anomaly monitoring system based on edge computing, comprising: The signal preprocessing module is used to acquire the original acoustic signal and perform layered noise reduction on the original acoustic signal to obtain a clean acoustic signal. The energy analysis module is used to calculate the frequency domain energy value of the clean acoustic signal to obtain energy density data, and calculate the energy difference based on the energy density data to obtain energy change data; The feature extraction module is used to perform risk analysis based on the energy change data, obtain the leakage risk frequency, and perform frequency domain feature decomposition based on the leakage risk frequency to obtain leakage feature components. The deviation detection module is used to calculate the energy deviation based on the leakage characteristic components, obtain the energy deviation sequence, and perform threshold comparison based on the energy deviation sequence to obtain the deviation anomaly sequence; The duration filtering module is used to track and statistically analyze the deviation anomaly sequence to obtain the duration of the anomaly, and to remove interference based on the duration of the anomaly to obtain the refined anomaly duration. The confidence assessment module is used to calculate the confidence level based on the duration of the refining anomaly, obtain the cumulative confidence level, and perform threshold comparison based on the cumulative confidence level to obtain the priority alarm sequence. The threshold correction module is used to acquire historical data of the safety valve, perform threshold correction based on the historical data of the safety valve to obtain a corrected alarm threshold, and compare the corrected alarm threshold with the priority alarm sequence to obtain an immediate alarm signal.

[0069] It should be noted that the edge computing-based safety valve status anomaly monitoring system provided in this embodiment of the invention is used to execute all the process steps of the edge computing-based safety valve status anomaly monitoring method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0070] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a safety valve status anomaly monitoring program based on edge computing. When the processor executes the computer program, it implements the steps described in the various embodiments of the safety valve status anomaly monitoring system based on edge computing, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the data preprocessing module.

[0071] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0072] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0073] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0074] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0075] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0076] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0077] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for monitoring the abnormal status of a safety valve based on edge computing, characterized in that, include: The original acoustic signal is acquired, and layered denoising is performed on the original acoustic signal to obtain a clean acoustic signal; The frequency domain energy value of the clean acoustic signal is calculated to obtain energy density data, and the energy difference is calculated based on the energy density data to obtain energy change data; Risk analysis is performed based on the energy change data to obtain the leakage risk frequency. Then, frequency domain feature decomposition is performed based on the leakage risk frequency to obtain leakage feature components. The energy deviation is calculated based on the leakage characteristic components to obtain an energy deviation sequence. A threshold comparison is then performed based on the energy deviation sequence to obtain an abnormal deviation sequence. The abnormal sequence is tracked and statistically analyzed to obtain the duration of the abnormality. Interference is then eliminated based on the duration of the abnormality to obtain the duration of the refined abnormality. The confidence level is calculated based on the duration of the refining anomaly to obtain the cumulative confidence level. The threshold is then compared based on the cumulative confidence level to obtain the priority alarm sequence. Historical data of the safety valve is acquired, and a threshold correction is performed based on the historical data to obtain a corrected alarm threshold. The corrected alarm threshold is then compared with the priority alarm sequence to obtain an immediate alarm signal.

2. The method for monitoring the abnormal status of a safety valve based on edge computing according to claim 1, characterized in that, The process of acquiring the original acoustic signal and performing layered noise reduction on the original acoustic signal to obtain a clean acoustic signal includes: The original acoustic signal is acquired, and wavelet decomposition is performed on the original acoustic signal to obtain a set of signal coefficients; Soft thresholding is performed on the set of signal coefficients to obtain filtered signal coefficients. The filtered signal coefficients are then combined in reverse to obtain a clean acoustic signal.

3. The method for monitoring the abnormal status of a safety valve based on edge computing according to claim 1, characterized in that, The step of calculating the frequency domain energy value of the clean acoustic signal to obtain energy density data, and calculating the energy difference based on the energy density data to obtain energy change data includes: Short-time Fourier transform is performed on the clean acoustic signal to obtain frequency domain amplitude data, and energy value is calculated based on the frequency domain amplitude data to obtain energy density data; Based on the energy density data, a spectrum is generated and logarithmic scaling is performed to obtain energy spectrum data; The energy difference between adjacent frames is calculated based on the energy spectrum data to obtain energy change data.

4. The method for monitoring the abnormal status of a safety valve based on edge computing according to claim 1, characterized in that, The risk analysis based on the energy change data yields the leakage risk frequency, and the frequency domain feature decomposition based on the leakage risk frequency yields leakage feature components, including: The energy change data is compared with a preset change threshold and marked to obtain a risk time map. The leakage risk frequency is obtained by comparing the risk time map with a preset risk model. Frequency domain features are extracted based on the leakage risk frequency to obtain preliminary feature components. These preliminary feature components are then weighted and filtered to obtain leakage feature components.

5. The method for monitoring the abnormal status of a safety valve based on edge computing according to claim 1, characterized in that, The step of calculating the energy deviation based on the leakage characteristic components to obtain an energy deviation sequence, and performing threshold comparison based on the energy deviation sequence to obtain an abnormal deviation sequence includes: The leakage characteristic component is compared with the preset low-temperature ambient noise baseline to obtain the energy deviation value. The deviation statistics within the preset sliding time window are calculated based on the energy deviation value to obtain the deviation window data. The deviation window data is compared with a preset deviation threshold. If the deviation window data exceeds the preset deviation threshold, it is marked to obtain a deviation anomaly sequence.

6. The method for monitoring the abnormal status of a safety valve based on edge computing according to claim 1, characterized in that, The step of tracking and statistically analyzing the deviation anomaly sequence to obtain the anomaly duration, and then performing interference removal based on the anomaly duration to obtain the refined anomaly duration, includes: The adjacent windows are merged according to the deviation anomaly sequence to obtain the merged anomaly sequence. The duration of the anomaly is obtained by time tracking and statistics based on the merged anomaly sequence. The duration of the anomaly is compared with the preset interference duration. If the duration of the anomaly is shorter than the preset interference duration, it is determined to be a non-steady-state fluctuation and is removed to obtain the refined anomaly duration.

7. The method for monitoring the abnormal status of a safety valve based on edge computing according to claim 1, characterized in that, The confidence level is calculated based on the duration of the refining anomaly to obtain a cumulative confidence level. A threshold comparison is then performed based on the cumulative confidence level to obtain a priority alarm sequence, including: A leakage change sequence is obtained by jointly constructing the leakage change sequence based on the refining anomaly duration and the leakage characteristic components. The confidence level is calculated based on the leakage change sequence to obtain a confidence level sequence. The cumulative confidence level data is obtained by integrating the confidence level sequence over time. The cumulative confidence data is compared with a preset confidence threshold. If the cumulative confidence data exceeds the preset confidence threshold, it is marked to obtain a priority alarm sequence.

8. The method for monitoring the abnormal status of a safety valve based on edge computing according to claim 1, characterized in that, The process of acquiring historical safety valve data, performing threshold correction based on the historical safety valve data to obtain a corrected alarm threshold, and comparing the corrected alarm threshold with the priority alarm sequence to obtain an immediate alarm signal includes: Obtain historical data of the safety valve, perform correction calculations based on the historical data of the safety valve to obtain a confidence correction factor, and perform threshold correction based on the confidence correction factor to obtain a corrected alarm threshold; The priority alarm sequence is compared with the corrected alarm threshold. If the deviation between the corrected alarm threshold and the priority alarm sequence is less than the preset alarm tolerance, an alarm message is generated to obtain an immediate alarm signal.

9. A safety valve status anomaly monitoring system based on edge computing, characterized in that, include: The signal preprocessing module is used to acquire the original acoustic signal and perform layered noise reduction on the original acoustic signal to obtain a clean acoustic signal. The energy analysis module is used to calculate the frequency domain energy value of the clean acoustic signal to obtain energy density data, and calculate the energy difference based on the energy density data to obtain energy change data; The feature extraction module is used to perform risk analysis based on the energy change data, obtain the leakage risk frequency, and perform frequency domain feature decomposition based on the leakage risk frequency to obtain leakage feature components. The deviation detection module is used to calculate the energy deviation based on the leakage characteristic components, obtain the energy deviation sequence, and perform threshold comparison based on the energy deviation sequence to obtain the deviation anomaly sequence; The duration filtering module is used to track and statistically analyze the deviation anomaly sequence to obtain the duration of the anomaly, and to remove interference based on the duration of the anomaly to obtain the refined anomaly duration. The confidence assessment module is used to calculate the confidence level based on the duration of the refining anomaly, obtain the cumulative confidence level, and perform threshold comparison based on the cumulative confidence level to obtain the priority alarm sequence. The threshold correction module is used to acquire historical data of the safety valve, perform threshold correction based on the historical data of the safety valve to obtain a corrected alarm threshold, and compare the corrected alarm threshold with the priority alarm sequence to obtain an immediate alarm signal.

Images