Method and device for detecting a leak in a pipe

Abstract

The application provides a pipeline leakage detection method, and relates to the field of pipeline leakage detection. The detection method comprises the following steps: acquiring a pressure signal data set of a pipeline; selecting Haar wavelet, Daubechies wavelet and Symlet wavelet as the base functions of discrete wavelet transform, and performing discrete wavelet transform on the pressure signal data set through the three kinds of wavelets respectively, the scale of the discrete wavelet transform of each wavelet is three, so as to obtain the decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform; inputting the decomposition coefficient sequences of the three kinds of discrete wavelet transforms into a self-attention mechanism model respectively, obtaining three kinds of pipeline leakage probabilities, and statistically analyzing the three kinds of pipeline leakage probabilities to obtain a conclusion of whether the pipeline leaks or not. The method can accurately determine whether the pipeline leaks or not through the analysis of the pressure signal by multiple wavelet transforms and a self-attention model for the detection of slight pipeline leakage.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of pipeline leakage detection, and provides a pipeline leakage detection method and device. BACKGROUND

[0002] Common pipeline leakage detection methods in the related art include negative pressure wave method, digital signal analysis method, optical fiber acoustic wave sensing method, etc. These detection methods can detect large-scale pipeline leakage to some extent, but for small-scale leakage (the area ratio of the leakage hole is less than 5%, and the leakage amount is less than 5% of the total flow), the detection accuracy is often difficult to meet the actual demand due to the weak leakage signal and the interference of background noise.

[0003] In recent years, with the development of artificial intelligence technology, deep learning models such as LSTM and Transformer have been gradually applied to the field of pipeline leakage detection. Although these models have certain advantages in processing complex data patterns, existing researches mainly focus on large-scale leakage detection, and the high-precision detection method for small-scale leakage is still imperfect. SUMMARY

[0004] Therefore, it is necessary to provide a pipeline leakage detection method and device to solve the above technical problems. The method can accurately determine whether the pipeline leaks by analyzing the pressure signal through multiple wavelet transforms and self-attention models.

[0005] In a first aspect, the present application provides a pipeline leakage detection method, which comprises:

[0006] obtaining a pressure signal data set of the pipeline;

[0007] selecting Haar wavelet, Daubechies wavelet and Symlet wavelet as the basis functions of discrete wavelet transform, and performing discrete wavelet transform on the pressure signal data set through the three kinds of wavelets respectively, and the discrete wavelet transform scale of each wavelet is three levels, so as to obtain the decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform;

[0008] inputting the decomposition coefficient sequences of the three kinds of discrete wavelet transforms into a self-attention mechanism model respectively to obtain three pipeline leakage probabilities, and performing statistics based on the three pipeline leakage probabilities to obtain a conclusion of whether the pipeline leaks based on the statistical result.

[0009] In one of the embodiments, the decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform includes: first, second and third level detail coefficient sequences, and third level approximation coefficient sequence;

[0010] The first-level detail coefficient sequence obtained by the discrete wavelet transform with the Haar wavelet as the base function is defined as a Haar wavelet first-level detail coefficient sequence, and the detection method further comprises adjusting the Haar wavelet first-level detail coefficient sequence.

[0011] In a case where the conclusion obtained based on the statistical result is that the pipeline is not leaking, the pressure signal data set is subjected to Haar wavelet transform to obtain a Haar wavelet first-level detail coefficient sequence, and a standard deviation of the Haar wavelet first-level detail coefficient sequence is calculated to obtain a first dynamic threshold value, a part of the Haar wavelet first-level detail coefficient sequence greater than the first dynamic threshold value is retained, and the rest is set to zero to obtain an adjusted Haar wavelet first-level detail coefficient sequence.

[0012] In one of the embodiments, the decomposition coefficient sequence corresponding to the discrete wavelet transform of the wavelet type comprises a first, a second and a third level detail coefficient sequence, and a third level approximation coefficient sequence;

[0013] The second-level detail coefficient sequence obtained by the discrete wavelet transform with the Daubechies wavelet as the base function is defined as a Daubechies wavelet second-level detail coefficient sequence, and the detection method further comprises adjusting the Daubechies wavelet second-level detail coefficient sequence, and the adjusting step comprises:

[0014] The coefficients in the Daubechies wavelet second-level detail coefficient sequence are subjected to standardization processing, each coefficient in the Daubechies wavelet second-level detail coefficient sequence after the standardization processing is traversed, if the absolute value of the coefficient is greater than an adaptive threshold value, the coefficient is adjusted to the deviation between the coefficient and the adaptive threshold value, the sign of the adjusted coefficient is the same as that of the unadjusted coefficient, and if the absolute value of the coefficient is less than or equal to the adaptive threshold value, the coefficient is set to zero;

[0015] The adaptive threshold value is obtained by multiplying the maximum absolute value of the coefficients in the Daubechies wavelet second-level detail coefficient sequence after the standardization processing by a preset proportion.

[0016] In one of the embodiments, the decomposition coefficient sequence corresponding to the discrete wavelet transform of the wavelet type comprises a first, a second and a third level detail coefficient sequence, and a third level approximation coefficient sequence;

[0017] The third-level detail coefficient sequence obtained by the discrete wavelet transform with the Symlet wavelet as the base function is defined as a Symlet wavelet third-level detail coefficient sequence, and the detection method further comprises adjusting the Symlet wavelet third-level detail coefficient sequence, and the adjusting step comprises:

[0018] Performing fast Fourier transform on the third-level detail coefficient sequence of the Symlet wavelet to obtain a complex spectrum, and calculating a power spectral density based on the complex spectrum;

[0019] Smoothing the power spectral density using a moving average filter, and selecting candidate peak values from the smoothed power spectral density, the candidate peak values being configured to satisfy: in the power spectral density, if the power spectral density of any target frequency point is greater than the power spectral densities of the adjacent frequency points on both sides, the target frequency point is taken as a candidate peak value;

[0020] Selecting a frequency point with the largest amplitude from the candidate peak values as a main frequency peak, calculating a half-width of the main frequency peak, and combining the frequency point, the amplitude and the half-width of the main frequency peak to obtain a combination feature of the power spectrum;

[0021] Splicing the third-level detail coefficient sequence of the Symlet wavelet and the combination feature to obtain an adjusted third-level detail coefficient sequence of the Symlet wavelet.

[0022] In one of the embodiments, the detection method further comprises:

[0023] Before performing the discrete wavelet transform on the pressure signal dataset, the pressure signal dataset is preprocessed as follows:

[0024] The pressure signal dataset is standardized, and the processed pressure signal dataset is segmented into time-series segments;

[0025] The time-series segments are input into a multi-layer perceptron for encoding adjustment, the encoded signals are spliced along the time dimension to obtain a global feature matrix, and principal component analysis is performed on the global feature matrix to reduce the data dimension to complete the preprocessing, so that the preprocessed pressure signal dataset is adapted to the discrete wavelet transform processing.

[0026] In one of the embodiments, the standardization of the pressure signal dataset comprises:

[0027] The mean and the standard deviation of the pressure signal dataset are calculated, the pressure time-series signals in the pressure signal dataset are subtracted by the mean of the pressure signal dataset, and the difference is divided by the standard deviation of the pressure signal dataset to obtain the standardized pressure time-series signals; and the standardized pressure time-series signals are summarized to obtain the standardized pressure signal dataset.

[0028] In one of the embodiments, the decomposition coefficient sequence corresponding to the wavelet type comprises: the first, second and third-level detail coefficient sequences, and the third-level approximation coefficient sequence, and the detection method further comprises:

[0029] The decomposition coefficient sequence of each discrete wavelet transform is first subjected to convolution processing, and then the convolution processing is input into the self-attention mechanism model to obtain the probability of pipeline leakage.

[0030] In one embodiment, when the first, second and third level detail coefficient sequences are subjected to convolution processing respectively, the convolution kernel is configured as 1*2.

[0031] In one embodiment, when the third level approximation coefficient sequence is subjected to convolution processing, the convolution kernel is configured as 1*1.

[0032] In a second aspect, the present application also provides a pipeline leakage detection device, which comprises:

[0033] A pressure detection unit configured to collect a pipeline pressure time series signal to obtain a pressure signal dataset of the pipeline;

[0034] A processing unit configured to apply the pipeline leakage detection method of the first aspect to perform pipeline leakage detection based on the pressure signal dataset, and output a conclusion of whether the pipeline leaks.

[0035] The pipeline leakage detection method described above can extract signal features at different frequencies and time scales by obtaining a pipeline pressure signal dataset and performing three-level discrete wavelet transform. Selecting Haar wavelet, Daubechies wavelet and Symlet wavelet as the basis function can capture characteristics such as sudden changes, noise and periodicity in the pressure signal, thereby comprehensively analyzing the signal. Inputting the decomposition coefficient sequence after transforming the basis function of each wavelet into the self-attention mechanism model can effectively capture the time dependence and global features in the signal. The self-attention mechanism identifies key features by calculating the correlation between different positions, thereby improving the detection accuracy. Since each wavelet basis function captures different features, combining the output results of them can reduce the error caused by a single basis function and increase the reliability of the detection results. This method takes advantage of multiple wavelets and self-attention mechanisms to effectively improve the accuracy and reliability of pipeline micro-leakage detection and reduce the risk of false positives and false negatives. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 Flowchart of the pipeline leakage detection method in one embodiment;

[0037] Figure 2 Flowchart of the discrete wavelet transform in one embodiment;

[0038] Figure 3 Flowchart of the self-attention mechanism model modeling in one embodiment;

[0039] Figure 4A flow chart of pre-processing of a pressure signal data set in one embodiment;

[0040] Figure 5 A pipeline leak detection device diagram in one embodiment. DETAILED DESCRIPTION

[0041] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0042] In one embodiment, as shown in Figure 1 A pipeline leak detection method is provided, which comprises the following steps:

[0043] Step 101: obtaining a pressure signal data set of a pipeline;

[0044] Specifically, professional data acquisition equipment can be used to collect pressure data, for example, a pressure sensor is installed on the pipeline to monitor the pipeline pressure signal in real time and store it as a pressure signal data set.

[0045] Step 102: selecting Haar wavelet, Daubechies wavelet and Symlet wavelet as the basis functions of discrete wavelet transform, and performing discrete wavelet transform on the pressure signal data set through the three kinds of wavelets respectively, and the discrete wavelet transform scale of each wavelet is three levels, to obtain the decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform;

[0046] Specifically, the pressure signal of the pipeline may contain characteristics of different frequencies, amplitudes and shapes due to the influence of many factors such as geology, environment, pipe material and leakage scale, and a single wavelet kernel is difficult to fully capture these characteristics. In view of this, the present application selects Haar wavelet, Daubechies wavelet and Symlet wavelet as the basis functions of discrete wavelet transform after in-depth analysis of the characteristics of pipeline pressure signal and the characteristics of wavelet kernel function, and multiple kernel wavelets are used for pipeline leak detection.

[0047] Specifically, after in-depth analysis of the characteristics of the pipeline pressure signal, on the one hand, the water supply pipeline leak signal often shows instantaneous mutation of pressure or flow, especially in the early stage of micro-leakage, the signal will produce steep edge characteristics due to the sudden drop of pipeline pressure. However, the traditional Fourier transform is difficult to capture such local transient changes, and the conventional wavelet kernel is easy to cause the mutation characteristics to be covered under the interference of high-frequency noise.

[0048] The Haar wavelet has a simple compactly supported orthogonal basis function (support length of 1), and its rectangular wave characteristics make it extremely sensitive to signal mutations (such as pressure drop), and can accurately locate the starting point of the leakage signal. In the case of a small leakage, the leakage signal amplitude is weak and easy to be overwhelmed by noise. The Haar wavelet can improve the accuracy of leakage positioning through its sharp time-domain response characteristics.

[0049] In a second aspect, the pipeline leakage signal usually contains complex background noise (such as pump vibration, water flow turbulence noise), which is mostly concentrated in the high frequency band, and overlaps with the transient high frequency characteristic band of the leakage signal, making it difficult for traditional noise reduction methods to distinguish effective leakage information.

[0050] The Daubechies (db4) wavelet has a 4th order vanishing moment, which can more smoothly approximate the low frequency trend of the signal, and its regularity can effectively suppress the pseudo Gibbs effect of high frequency noise, avoiding signal distortion after noise reduction. In view of the problem that the high frequency noise and the leakage signal band overlap, the Daubechies wavelet avoids signal detail loss and improves the detection rate of small leakage signals through its frequency band segmentation characteristics and high order vanishing moment.

[0051] In a third aspect, the leakage signal of the water supply pipeline may exhibit certain symmetry or quasi-periodic characteristics (such as decaying oscillation of pressure waveform) due to factors such as fluid dynamics, periodic start-stop of the pump station, or resonance of the pipeline structure. However, the non-symmetry of conventional wavelet kernels (such as db series) can cause phase shift, affecting the integrity of the leakage characteristics.

[0052] The Symlet wavelet (sym5) is an improved version of the Daubechies wavelet, which has higher symmetry by optimizing the filter coefficients. In view of the detection requirement of the symmetry component in the leakage signal, the sym5 wavelet solves the phase distortion problem caused by the non-symmetry wavelet kernel through its symmetry design, so that the model can more accurately capture the oscillation characteristics of the leakage signal, and improve the detection robustness under complex conditions.

[0053] In summary, the multi-wavelet cooperative mechanism proposed in this application can accurately locate the leakage mutation through the Haar wavelet, suppress high frequency noise through the db4 wavelet, and retain the symmetry characteristics through the sym5 wavelet, realizing multi-dimensional analysis of small leakage signals. The synergistic effect of the three not only solves the limitations of a single wavelet kernel (such as Haar sensitivity to noise and db4 phase shift), but also covers the complex characteristics of the leakage signal (mutation, noise, periodicity) through the complementarity of multiple kernels, ultimately building a high-precision, strong-robustness leakage detection model.

[0054] Further, the pressure signal dataset is respectively subjected to discrete wavelet transform by Haar wavelet, Daubechies wavelet and Symlet wavelet, and the transform scale of each wavelet is set to three levels (i.e., three-level decomposition is performed on each wavelet). The three-level decomposition means that the pressure signal is gradually decomposed into sub-band signals of different frequency bandwidths, so as to obtain more detailed signal features. Through three-level decomposition, a decomposition coefficient sequence corresponding to each wavelet type can be obtained, which includes an approximation coefficient and a detail coefficient. The approximation coefficient can reflect the low-frequency component and overall trend of the pressure signal, and the detail coefficient can capture the high-frequency component and local change of the pressure signal. The decomposition coefficient sequence corresponding to each wavelet type can comprehensively reflect the features of the pressure signal at multiple levels.

[0055] The flow of discrete wavelet transform is shown in Figure 2 . The encoded signal is respectively subjected to Haar wavelet, Daubechies wavelet and Symlet wavelet transform, to obtain a decomposition coefficient sequence corresponding to each wavelet type. Discrete wavelet transform (DWT) is a signal processing technology that can analyze the local features of a signal at multiple scales, and is widely used in fields such as signal compression, denoising and feature extraction. Assuming that the time series is x [ n ], the low-pass filter coefficient of the wavelet basis function is h [ m ], and the high-pass filter coefficient is g [ m ]. Then, the discrete wavelet transform is represented as follows:

[0056] First-level decomposition:

[0057] First-level approximation coefficient (low-frequency component):

[0058] .

[0059] First-level detail coefficient (high-frequency component):

[0060] .

[0061] wherein cA 1[ k ] is the first approximation coefficient after first-level decomposition; cD 1[ k ] is the first-level detail coefficient after first-level decomposition; h [ m ] is the low-pass filter coefficient; g [ m ] is the high-pass filter coefficient; x [ n ] is the original time series; KThe new time index, due to the effect of downsampling, has a value range of half the original signal length; M This is the filter index, used to iterate through the filter coefficients.

[0062] For multi-level decomposition, the approximation coefficients are decomposed recursively. For example, a two-level decomposition:

[0063] Approximation coefficient for the second season:

[0064] .

[0065] Second level of detail:

[0066] .

[0067] And so on, we obtain the j-th level decomposition. cA j [ k ]and cD j [ k ].

[0068] in, cA j [ k ] is the first j The approximate coefficients after level decomposition reflect the low-frequency components of the pressure signal at this scale. k It is a new time index, and due to the effect of downsampling, its value range is half the length of the original signal. cD j [ k ] is the first j The detail coefficients after level decomposition reflect the high-frequency components of the pressure signal at that scale. Similarly, due to the effect of downsampling, their value range is half the length of the original signal.

[0069] It should be noted that the input signal is subjected to Haar wavelet, Daubechies wavelet, and Symlet wavelet transforms respectively. Each wavelet transform performs a three-level decomposition. Taking the three-level decomposition of the Haar wavelet as an example, the specific implementation steps are as follows:

[0070] First-level decomposition: Input is Z[n], output is the first-level approximation coefficient cA 11 and detail coefficient cD 11 , and the first level of detail coefficients cD 11 Recorded as D11;

[0071] Second-level decomposition: Recursive decomposition of cA 11 Output cA 21 and cD 21 , and the second level of detail coefficients cD 21 Recorded as D21;

[0072] Third-level decomposition: recursive decomposition of cA 21 , output cA 31 and cD 31 , the third-level detail coefficient cD 31 is denoted as D31, and the third-level approximation coefficient cA 31 is denoted as A31.

[0073] Similarly, the first-level detail coefficient D12, the second-level detail coefficient D22, the third-level detail coefficient D32 and the third-level approximation coefficient A32 corresponding to the Daubechies wavelet can be obtained; and the first-level detail coefficient D13, the second-level detail coefficient D23, the third-level detail coefficient D33 and the third-level approximation coefficient A33 corresponding to the Symlet wavelet can be obtained.

[0074] By accurately positioning the leakage mutation through the Haar wavelet, suppressing high-frequency noise through the Daubechies wavelet, and retaining the symmetry feature through the Symlet wavelet, multi-dimensional analysis of the micro leakage signal is realized. The synergistic effect of the three solves the limitations of a single wavelet kernel (such as Haar sensitivity to noise and Daubechies phase shift), and covers the complex characteristics (mutation, noise, periodicity) of the leakage signal through multi-kernel complementarity.

[0075] With the step-by-step decomposition of DWT (Discrete Wavelet Transform), the frequency characteristics, time resolution and signal characteristics of the detail coefficients also change.

[0076] Frequency characteristics: with the increase of the decomposition level, the frequency range corresponding to the detail coefficient gradually decreases, gradually covering different frequency components of the signal from high frequency to low frequency.

[0077] Time resolution: the lower the level of the detail coefficient, the higher the time resolution, which can more accurately locate the transient change in the signal; the higher the level of the detail coefficient, the lower the time resolution, but it can reflect the macroscopic change trend of the signal.

[0078] Signal characteristics: different levels of detail coefficients reflect different scale characteristics in the signal. The low-level detail coefficient mainly captures the rapid change and high-frequency characteristics of the signal, while the high-level detail coefficient captures the slow change and low-frequency characteristics of the signal. The target of the present application is to detect micro leakage. As can be known from the foregoing signal characteristic analysis part, the low-level detail coefficient can fully reflect this leakage characteristic. After the present application attempts to set the decomposition level to 1, 2, 3, 4 and 5 and checks the final detection effect, it is found that the detection accuracy almost no longer has an upward trend after the decomposition level exceeds 3 levels, and therefore the present application sets the decomposition level to 3, that is, 3-level decomposition is performed, and finally 1 approximation coefficient sequence and 3 detail coefficient sequences are obtained, that is, Figure 2A3 (approximation coefficient) and D1, D2, D3 (detail coefficients) in the wavelet transform.

[0079] In addition, each wavelet transform has different characteristics in its detail coefficients. In order to comprehensively coordinate each wavelet transform and achieve better recognition effect, the application adjusts part of the coefficients, and the specific adjustment steps are described later.

[0080] Step 103: input the decomposition coefficient sequences of the three discrete wavelet transforms into the self-attention mechanism model respectively, obtain the probabilities of the three pipeline leakages, and count the pipeline leakage probabilities output by the three, and obtain the conclusion of whether the pipeline leaks based on the statistical results.

[0081] Specifically, when detecting pipeline leakage, three different wavelet transforms are used, each of which can decompose complex pressure signals into detail coefficient sequences of different frequencies and an approximation coefficient sequence. Further, the decomposition coefficient sequence of each wavelet transform is input into the self-attention mechanism model. The self-attention mechanism can analyze the relationship between each element and other elements in the sequence data, thereby identifying which part is most likely to indicate the occurrence of leakage.

[0082] The self-attention mechanism model can calculate a probability value according to the input decomposition coefficient sequence, which can represent the possibility of pipeline leakage. The probability values obtained by the three wavelet transforms are statistically analyzed, and the probabilities of leakage in different frequency bands are considered comprehensively. If most of the probability values point to leakage, it can be considered that the pipeline is likely to leak. Exemplarily, the probability values obtained by the three wavelet transforms can be averaged, or weighted average processing can be selected.

[0083] The modeling process of the self-attention mechanism model is shown in Figure 3 After multi-core wavelet collaborative multi-scale decomposition, that is, 3-level decomposition of Haar wavelet, Daubechies wavelet and Symlet wavelet, the first-level detail coefficient, the second-level detail coefficient, the third-level detail coefficient and the third-level approximation coefficient corresponding to each wavelet are obtained, and each coefficient is subjected to convolution operation, that is, input into four convolution modules Conv1, Conv2, Conv3 and Conv4 for further adjustment. Then, the output signal after dynamic adjustment is input into the self-attention mechanism model, and the self-attention mechanism of the self-attention mechanism model is used to capture the long-distance correlation of the leakage signal (such as the propagation characteristics of the pressure fluctuation caused by leakage) from a global perspective, realizing long-time sequence dependence modeling.

[0084] The convolution in the embodiments of the present application is designed as a learnable convolution, which can be automatically adjusted according to the characteristics of different signals during model training. For example, taking Haar wavelet transform as an example, the detail coefficients (D11-D31) after wavelet transform are processed by convolution kernel 1x2, and the approximation coefficient A31 is processed by convolution kernel 1x1, and the implementation steps are as follows:

[0085] For the detail coefficients (D11-D31), a 1x2 convolution kernel is set and randomly initialized, the step is 1, and the output channel is 32. For the approximation coefficient (A31), a 1x1 convolution kernel is set and randomly initialized, the step is 1, and the output channel is 32. The activation function is ReLU. Similarly, Daubechies wavelet and Symlet wavelet as base function can also be calculated by the above method. By designing the convolution kernel of the detail coefficient as 1x2, the influence of noise on the most core feature information contained in the detail information in the leakage detection can be reduced. Moreover, since the approximation coefficient reflects the stationary part of the signal, configuring the convolution kernel of the approximation coefficient as 1x1 can reduce the parameter amount learned by the subsequent self-attention mechanism model and improve the calculation efficiency. Through the above method, the model parameters can be dynamically adjusted according to the characteristics of the signal, realizing dynamic feature optimization of the model, self-adaptive distinguishing of noise and effective leakage features, and avoiding signal distortion caused by fixed threshold or traditional filtering.

[0086] Further, the signals after 3 kinds of wavelet transform after convolution processing are input into the self-attention mechanism model for time series modeling, capturing the long-distance correlation of the leakage signal from a global perspective (such as the pressure fluctuation propagation characteristics caused by leakage), breaking through the short-term memory limitation of traditional RNN model caused by gradient disappearance, and realizing long-time sequence dependence modeling. The output of the self-attention mechanism model is the detection result, which is given in the form of probability, and the implementation steps are as follows:

[0087] For the 3 kinds of wavelet transform, 3 self-attention mechanism models are set respectively (such as Figure 3 ), the number of heads in the multi-head self-attention mechanism model is set to 8, and the hidden layer dimension is set to 512. The probabilities output by the 3 self-attention mechanism models are averaged to obtain the final detection result, i.e. the probability of pipeline leakage.

[0088] According to the above description, through the collaborative design of the learnable convolution operation and the Transformer, the precision bottleneck problem caused by noise interference and insufficient time sequence dependence modeling in the traditional method in the micro leakage detection is solved. Moreover, three independent Transformers (self-attention mechanism) are adopted in the present application, so that the overall algorithm can fully utilize the good adaptability, flexibility, robustness and long sequence processing capability of the Transformer for time sequence signals, and on the other hand, each Transformer model can be optimized for the data decomposed by a specific wavelet kernel function, so as to improve the learning effect of specific features and further improve the robustness and accuracy of classification. The three classification probabilities output by the model are averaged to obtain the final detection result.

[0089] In one embodiment, as shown in FIG. 1, before performing the discrete wavelet transform on the pressure signal data set, the pressure signal data set is preprocessed as follows: Figure 4

[0090] Step 401: standardizing the pressure signal data set, and dividing the processed pressure signal data set into time series segments;

[0091] Specifically, the mean and standard deviation of all pressure time series signals in the pressure signal data set are calculated. The mean can represent the average level of the pressure signal data, and the standard deviation can reflect the dispersion degree of the pressure signal data. Further, for each pressure time series signal in the pressure data set, each value in it is subtracted by the mean of the entire data set, so that the pressure signal data set is centered around zero mean. The difference value is divided by the standard deviation of the pressure signal data set to obtain the standardized pressure time series signal.

[0092] For example, assume that the water supply pipeline pressure time series signal x[n] has a sampling frequency of 200 Hz and a length of N=12000 (1 minute of data).

[0093] Data standardization: x'[n]=(x[n]-μ) / σ.

[0094] Where μ is the mean and σ is the standard deviation.

[0095] After completing the standardization process, it is divided into time series segments, thereby converting the pressure signal data set into an input format suitable for a multilayer perceptron. The sliding window method can be used for segmentation. For example, the window length can be set to 512 data points, and the step length can be set to 256 data points. Starting from the beginning of the pressure signal data set, a data segment with a length of 512 is intercepted as a time series segment. The window is moved by the set step length of 256, and the subsequent segments are sequentially intercepted until the entire data set is processed, generating 46 time series segments. ​

[0096] Step 402: input the time-series segments into a multi-layer perceptron for encoding adjustment, and splice the encoded signals along the time dimension to obtain a global feature matrix, and perform data dimension reduction on the global feature matrix using principal component analysis to complete preprocessing, so that the preprocessed pressure signal dataset is adapted to discrete wavelet transform processing.

[0097] Specifically, the time-series segments are input into a multi-layer perceptron (MLP) for encoding adjustment. As a kind of feedforward neural network, the multi-layer perceptron can learn complex patterns and features in the data through nonlinear transformation of its hidden layers. The encoded signals, i.e., the output after MLP processing, are spliced along the time dimension to form a global feature matrix that can integrate the feature information of all time-series segments.

[0098] To further reduce the dimension of the pressure signal data and remove redundant information, principal component analysis (PCA) can be used to reduce the dimension of the global feature matrix. PCA projects the original data into a low-dimensional space by finding the main variation direction, i.e., the principal component, while preserving as much important information as possible. Further, the pressure signal dataset can be converted into a dataset that preserves key features and reduces dimension, making it compatible with discrete wavelet transform processing. Discrete wavelet transform can decompose signals at multiple scales, extracting features at different frequencies and time scales. The preprocessed pressure signal dataset can provide clearer and more focused input for wavelet transform, helping to improve the accuracy and reliability of subsequent pipeline leak detection.

[0099] Exemplarily, the network structure of the MLP encoder is set as follows:

[0100] Input layer: 512 neurons (corresponding to window length);

[0101] Hidden layer 1: 256 neurons, activation function ReLU;

[0102] Hidden layer 2: 128 neurons, activation function ReLU;

[0103] Dropout: inter-layer dropout rate 0.2, to prevent overfitting;

[0104] Output layer: 64 neurons, linear activation, output encoding Z ∈ R 46×64 ;

[0105] Splice along the time dimension to generate a global feature matrix Z global ∈ R 2944 , compress the dimension to 256 by PCA, and adapt to subsequent wavelet decomposition.

[0106] In one embodiment, the first-level detail coefficient sequence obtained by the discrete wavelet transform with the Haar wavelet as the base function is defined as the Haar wavelet first-level detail coefficient sequence, and the detection method further comprises adjusting the Haar wavelet first-level detail coefficient sequence. The adjusting step comprises:

[0107] In the case where the conclusion obtained based on the statistical result is that the pipeline is not leaking, the pressure signal data set is subjected to wavelet transform and the standard deviation is calculated, the first dynamic threshold is obtained based on the standard deviation, the part of the Haar wavelet first-level detail coefficient sequence that is greater than the first dynamic threshold is retained, and the rest is set to zero, so as to obtain the adjusted Haar wavelet first-level detail coefficient sequence.

[0108] Specifically, in the case where the conclusion obtained based on the statistical result is that the pipeline is not leaking, the pressure signal data set can be decomposed using the Haar wavelet, and the first-level detail coefficient sequence is calculated. Further, the standard deviation of the pressure signal data set is calculated. The standard deviation can reflect the dispersion degree of the pressure data and is an important basis for determining the dynamic threshold. The dynamic threshold can be determined based on the calculated standard deviation and is usually set to several times of the standard deviation, for example, three times of the standard deviation. Such setting can effectively filter out most of the noise signals while retaining the signal part that may contain leakage characteristics.

[0109] The calculated dynamic threshold is applied to the Haar wavelet first-level detail coefficient sequence, and the part of the first-level detail coefficient sequence that is greater than the dynamic threshold is retained, and the rest is set to zero.

[0110] Exemplarily, for the Haar wavelet first-level detail coefficient sequence D1 obtained by the Haar wavelet transform, the following adjustment is made, and the steps are as follows:

[0111] The first dynamic threshold θ is set to 3σ n , and σ n is the standard deviation of the non-leakage signal.

[0112] When the pipeline is normally operated without leakage, 10 minutes of pressure signals are collected and subjected to Haar wavelet transform, the Haar wavelet first-level detail coefficient sequence is obtained, and the standard deviation of the Haar wavelet first-level detail coefficient sequence is calculated.

[0113] The standard deviation is calculated as follows:

[0114] .

[0115] Wherein, N is the number of pressure signal samples when the pipeline is normally operated without leakage; x i is the i-th Haar wavelet first-level detail coefficient; and μ is the mean value of the Haar wavelet first-level detail coefficient sequence when the pipeline is operated without leakage.

[0116] The part of the first-level detail coefficient sequence of Haar wavelet larger than the first dynamic threshold is reserved, and the rest is set to zero, and the calculation formula is as follows:

[0117] .

[0118] In the embodiment of the application, the dynamic threshold θ = 3σ n .

[0119] In one embodiment, the second-level detail coefficient sequence obtained by performing discrete wavelet transform with Daubechies wavelet as the base function is defined as the Daubechies wavelet second-level detail coefficient sequence, and the detection method further includes adjusting the Daubechies wavelet second-level detail coefficient sequence, and the adjusting step includes:

[0120] The coefficients in the Daubechies wavelet second-level detail coefficient sequence are standardized, and each coefficient in the standardized Daubechies wavelet second-level detail coefficient sequence is traversed, if the absolute value of the coefficient is greater than the adaptive threshold, the coefficient is adjusted to the deviation between the coefficient and the adaptive threshold, the sign of the adjusted coefficient is the same as that before adjustment, and if the absolute value of the coefficient is less than or equal to the adaptive threshold, the coefficient is set to zero;

[0121] The adaptive threshold is obtained by multiplying the maximum absolute value of the coefficients in the standardized Daubechies wavelet second-level detail coefficient sequence by a preset proportion.

[0122] Specifically, each coefficient in the Daubechies wavelet second-level detail coefficient sequence is standardized. The difference between each coefficient and the mean value of the sequence is calculated, and it is divided by the standard deviation of the sequence to realize the conversion of the coefficient into a standardized form with zero mean and unit variance. In this way, the influence of the dimension can be eliminated, and the data is more comparable.

[0123] Each coefficient in the standardized Daubechies wavelet second-level detail coefficient sequence is traversed to determine the relationship between its absolute value and the adaptive threshold. The adaptive threshold is determined by taking the maximum absolute value of the coefficients in the standardized sequence and multiplying it by a preset proportion. The preset proportion can be a parameter determined by experience or experiment, which is used to control the strictness of the threshold.

[0124] Exemplarily, for the second-level detail coefficient D22 in the second-level detail coefficient sequence after Daubechies decomposition, considering that the D2 contains mid-frequency band information, soft threshold denoising adjustment is performed on it, so as to retain signal continuity while denoising, and the steps are as follows:

[0125] The coefficient sequence in the second-level detail coefficient sequence after the Daubechies wavelet two-level decomposition is D22={d1, d2,...,d N} with a length of N; the D22 is normalized, and the calculation formula is as follows:

[0126] .

[0127] Wherein μ is the mean of the coefficients in the second-level detail coefficient sequence, and σ is the standard deviation of the coefficients in the second-level detail coefficient sequence.

[0128] An adaptive threshold is set based on the pressure data signal strength:

[0129] λ=0.1·max|D22|.

[0130] Each coefficient d k in D22 is adjusted as follows:

[0131] If the absolute value of the coefficient is greater than the adaptive threshold, the coefficient is adjusted to the deviation between the coefficient and the adaptive threshold, and the sign of the adjusted coefficient is the same as that of the unadjusted coefficient.

[0132] That is, if |d k |>λ, the contracted value d k ' is calculated as follows: d k '=sign(d k )⋅(∣d k ∣−λ).

[0133] If the absolute value of the coefficient is less than or equal to the adaptive threshold, the coefficient is set to zero: d k '=0.

[0134] Wherein sign is a sign function, which is defined as follows:

[0135] .

[0136] In one embodiment, the third-level detail coefficient sequence obtained by defining the Symlet wavelet as the base function for discrete wavelet transform is defined as the Symlet wavelet third-level detail coefficient sequence, and the detection method further comprises adjusting the Symlet wavelet third-level detail coefficient sequence, and the adjusting step comprises:

[0137] The Symlet wavelet third-level detail coefficient sequence is subjected to fast Fourier transform to obtain a complex frequency spectrum, and the power spectral density is calculated based on the complex frequency spectrum;

[0138] The power spectral density is smoothed using a moving average filter, and candidate peaks are selected from the smoothed power spectral density, the candidate peaks being configured to satisfy: in the power spectral density, if the power spectral density of any target frequency point is greater than the power spectral densities of the adjacent frequency points on both sides, the target frequency point is taken as a candidate peak;

[0139] The frequency point with the largest amplitude is selected as the main frequency peak from the candidate peaks, the half-height width of the main frequency peak is calculated, and the frequency point, amplitude and half-height width of the main frequency peak are combined to obtain a combined feature of the power spectrum;

[0140] The third-level detail coefficient sequence of the Symlet wavelet is spliced with the combined feature to obtain an adjusted third-level detail coefficient sequence of the Symlet wavelet.

[0141] Specifically, the third-level detail coefficient sequence of the Symlet wavelet is subjected to fast Fourier transform (FFT) to convert the time domain signal into a frequency domain signal, and a complex frequency spectrum is obtained. The power spectral density is calculated based on the complex frequency spectrum, which represents the power distribution of the signal at different frequencies.

[0142] Further, the power spectral density is smoothed using a moving average filter to reduce noise fluctuations and make the power spectral density curve smoother. Then, candidate peaks are selected from the smoothed power spectral density. In the power spectral density, if the amplitude of any target frequency point is greater than the amplitudes of the adjacent frequency points on both sides, the target frequency point is selected as a candidate peak. The frequency point with the largest amplitude is selected as the main frequency peak from all candidate peaks. The half-height width of the main frequency peak is calculated, that is, the interval between the two frequency points on both sides of the main frequency peak whose power spectral densities are equal to half the amplitude of the main peak is found. The frequency point, amplitude and half-height width of the main frequency peak are combined into a feature vector, which is referred to as a combined feature of the power spectrum.

[0143] The third-level detail coefficient sequence of the Symlet wavelet is spliced with the combined feature, that is, the combined feature is attached to the original detail coefficient sequence to form a new sequence, which is an adjusted third-level detail coefficient sequence of the Symlet wavelet.

[0144] Exemplarily, for the data after the Symlet wavelet transform, the third-level detail coefficient D33 of the Symlet wavelet is adjusted to highlight the medium and low frequency characteristics (i.e. trend changes and periodic characteristics) of the signal, and the steps are as follows:

[0145] The third-level detail coefficient D33 of the Symlet wavelet is subjected to fast Fourier transform (FFT) to obtain a complex frequency spectrum S(f); and the power spectral density P(f) thereof is calculated, and the calculation formula is as follows:

[0146] .

[0147] Wherein, Fs is sampling frequency, N represents sampling quantity.

[0148] The power spectrum density is smoothed by using a moving average filter (the window length is set to 5 in the embodiment of the application) to reduce noise fluctuation, and the calculation formula is as follows:

[0149] .

[0150] If the following condition is met, f is marked as a candidate peak value:

[0151] P smooth (f)>P smooth (f−1) and P smooth (f)>P smooth (f+1).

[0152]

[0153] The dynamic threshold θ is set to μ+2σ, wherein μ is the mean of the power spectrum density, and σ is the standard deviation of the power spectrum density

[0154] Further, the candidate peak values satisfying P smooth (f)>θ are reserved, the peak with the largest amplitude is selected as the main frequency f peak , and the amplitude P peak =P(f peak ) is recorded.

[0155] The frequency points f peak and f left satisfying P(f)=0.5P right are found on both sides of the main frequency peak, and the calculation formula of the half-height width FWHM is as follows:

[0156] FWHM=f right −f left .

[0157] The combined features of the power spectrum are Feature=[f peak ,P peak ,FWHM], and the final result is [D33, Feature] by splicing D33 and the combined features.

[0158] According to the above description, in the embodiment of the application, the partial detail coefficients obtained by the three wavelet transforms are adjusted accordingly, so as to highlight the advantages and characteristics of each wavelet transform, and realize the goal of comprehensive cooperation and complementary characteristics of multi-core wavelet.

[0159] ​To further illustrate the detection method provided by the embodiments of the present application, more than 3000 pipeline signals with a duration of 1-2 minutes and a leakage phenomenon are used as negative samples, and more than 6000 pipeline signals with a duration of 1-2 minutes and no leakage phenomenon are used as positive samples in the embodiments of the present application. The signals are used to train the model, and the model is evaluated by using three indexes of precision P, recall R and F1 value. The precision P is the proportion of the real positive samples predicted as positive samples to all the positive samples, the recall R is the proportion of the real positive samples predicted as positive samples to the actual positive samples, and the F1 value considers the two evaluation indexes of precision and recall, and the range is [0, 1]. The calculation formulas of the evaluation indexes are as follows:

[0160] ;

[0161] ;

[0162] ;

[0163] In the formula, TP is the real positive sample and the predicted positive sample, FP is the real negative sample and the predicted positive sample, FN is the real positive sample and the predicted negative sample, and TN is the real negative sample and the predicted negative sample.

[0164] The effect of the method provided by the present application and the classical support vector machine SVM is shown in the following table. It is found from the table that the accuracy of the leakage signal recognition of the present application reaches 97%, which better achieves the design purpose.

[0165]

[0166] In summary, the leakage detection method provided by the embodiments of the present application has at least the following advantages: a multi-kernel wavelet cooperative mechanism is proposed, the limitations of a single wavelet kernel are solved, and the leakage signal mutation, noise, periodicity and other complex characteristics are covered through the complementarity of multiple kernels. A dynamic feature optimization mechanism is designed and deeply fused with the time sequence modeling capability of the Transformer, which not only solves the “feature extraction difficulty” problem of the micro leakage signal caused by the weak amplitude and noise, but also makes up for the deficiency of the traditional method in modeling the leakage propagation effect through global time sequence analysis.

[0167] Based on the same idea, as shown in Figure 5 , the present application also provides a pipeline leakage detection device, characterized in that the pipeline leakage detection device comprises:

[0168] The pressure detection unit 501 is configured to collect the pipeline pressure time sequence signal and obtain the pressure signal data set of the pipeline;

[0169] The processing unit 502 is configured to apply the above-mentioned pipeline leakage detection method to detect pipeline leakage based on the pressure signal data set, and output a conclusion of whether the pipeline leaks.

[0170] Specifically, the function of the pressure detection unit 501 can be to collect the time series signal of the pressure in the pipeline and integrate to form a pressure signal data set, providing basic data for subsequent analysis. The processing unit 502 can be responsible for analyzing the data set using the above-mentioned pipeline leakage detection method, including steps such as multi-wavelet transform, convolution processing, and self-attention mechanism model, and finally outputting a conclusion of whether the pipeline leaks.

[0171] The pipeline leakage detection device can realize the organic combination of data acquisition and intelligent analysis, and can efficiently and accurately detect pipeline leakage, especially suitable for the detection scene of small leakage.

[0172] The technical features of the above embodiments can be combined in any way. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0173] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of protection of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method for detecting pipeline leaks, characterized in that, The detection method includes the following steps: Acquire the pressure signal dataset of the pipeline; Haar wavelet, Daubechies wavelet and Symlet wavelet are selected as the basis functions of discrete wavelet transform. Discrete wavelet transform is performed on the pressure signal dataset by these three wavelets respectively. The discrete wavelet transform scale of each wavelet is three levels to obtain the decomposition coefficient sequence of the discrete wavelet transform of the corresponding wavelet type. The decomposition coefficient sequences of three discrete wavelet transforms are input into the self-attention mechanism model to obtain three pipeline leakage probabilities. Statistics are performed on the three pipeline leakage probabilities, and a conclusion on whether the pipeline is leaking is obtained based on the statistical results.

2. The method for detecting pipeline leakage according to claim 1, characterized in that, The decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform includes: first, second and third level detail coefficient sequences, and third level approximation coefficient sequence; The first-level detail coefficient sequence obtained by performing discrete wavelet transform using Haar wavelets as basis functions is defined as the Haar wavelet first-level detail coefficient sequence. The detection method further includes adjusting the Haar wavelet first-level detail coefficient sequence, and the adjustment steps include: If the conclusion based on statistical results is that the pipeline is not leaking, a Haar wavelet transform is performed on the pressure signal dataset to obtain the first-level detail coefficient sequence of the Haar wavelet. The standard deviation of the first-level detail coefficient sequence of the Haar wavelet is calculated to obtain a first dynamic threshold. The portion of the first-level detail coefficient sequence of the Haar wavelet that is greater than the first dynamic threshold is retained, and the remaining portion is set to zero to obtain an adjusted first-level detail coefficient sequence of the Haar wavelet.

3. The method for detecting pipeline leakage according to claim 1, characterized in that, The decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform includes: first, second and third level detail coefficient sequences, and third level approximation coefficient sequence; The second-level detail coefficient sequence obtained by performing discrete wavelet transform using Daubechies wavelet as a basis function is defined as the Daubechies wavelet second-level detail coefficient sequence. The detection method further includes adjusting the Daubechies wavelet second-level detail coefficient sequence, and the adjustment steps include: The coefficients in the second-level detail coefficient sequence of the Daubechies wavelet are standardized. Each coefficient in the standardized second-level detail coefficient sequence of the Daubechies wavelet is traversed. If the absolute value of the coefficient is greater than the adaptive threshold, the coefficient is adjusted to the deviation between the coefficient and the adaptive threshold. The sign of the adjusted coefficient is the same as that before the adjustment. If the absolute value of the coefficient is less than or equal to the adaptive threshold, the coefficient is set to zero. The adaptive threshold is obtained by multiplying the maximum absolute value of the coefficients in the second-level detail coefficient sequence of the Daubechies wavelet after standardization by a preset ratio.

4. The method for detecting pipeline leakage according to claim 1, characterized in that, The decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform includes: first, second and third level detail coefficient sequences, and third level approximation coefficient sequence; The third-level detail coefficient sequence obtained by performing discrete wavelet transform using Symlet wavelet as the basis function is defined as the Symlet wavelet third-level detail coefficient sequence. The detection method further includes adjusting the Symlet wavelet third-level detail coefficient sequence, and the adjustment steps include: A fast Fourier transform is performed on the Symlet wavelet third-level detail coefficient sequence to obtain a complex spectrum, and the power spectral density is calculated based on the complex spectrum. The power spectral density is smoothed using a moving average filter, and candidate peaks are selected from the smoothed power spectral density. The candidate peaks are configured to satisfy the following condition: if the power spectral density of any target frequency point is greater than the power spectral density of the two adjacent frequency points, then the target frequency point is selected as the candidate peak. Select the frequency point with the largest amplitude from the candidate peaks as the main frequency peak, calculate the full width at half maximum (FWHM) of the main frequency peak, and combine the frequency point, amplitude, and FWHM of the main frequency peak to obtain the combined characteristics of the power spectrum. The Symlet wavelet third-level detail coefficient sequence is concatenated with the combined feature to obtain the adjusted Symlet wavelet third-level detail coefficient sequence.

5. The method for detecting pipeline leakage according to any one of claims 1 to 4, characterized in that, The detection method further includes: Before performing discrete wavelet transform on the pressure signal dataset, the pressure signal dataset is preprocessed as follows: The pressure signal dataset is standardized, and the processed pressure signal dataset is segmented into time-series fragments. The time-series segments are input into a multilayer perceptron for encoding and adjustment. The encoded signals are spliced ​​along the time dimension to obtain a global feature matrix. Principal component analysis is used to perform data dimensionality reduction on the global feature matrix to complete the preprocessing, so that the preprocessed pressure signal dataset is compatible with discrete wavelet transform processing.

6. The method for detecting pipeline leakage according to claim 5, characterized in that, The standardization process for the pressure signal dataset includes: Calculate the mean and standard deviation of the pressure signal dataset, subtract the mean of the pressure signal dataset from the pressure time series signals in the pressure signal dataset, and divide the difference by the standard deviation of the pressure signal dataset to obtain the standardized pressure time series signals; summarize the standardized pressure time series signals to obtain the standardized pressure signal dataset.

7. The method for detecting pipeline leakage according to claim 1, characterized in that, The decomposition coefficient sequence of the corresponding wavelet type discrete wavelet transform includes: first, second, and third level detail coefficient sequences, and a third level approximation coefficient sequence. The detection method further includes: First, the decomposition coefficient sequence of each discrete wavelet transform is convolved, and then input into the self-attention mechanism model to obtain the probability of pipeline leakage.

8. The method for detecting pipeline leakage according to claim 7, characterized in that, When performing convolution processing on the first, second, and third level detail coefficient sequences respectively, the convolution kernel is configured to be 1*2.

9. The method for detecting pipeline leakage according to claim 7, characterized in that, When performing convolution on the third-level approximation coefficient sequence, the convolution kernel is configured as 1*1.

10. A pipeline leak detection device, characterized in that, The pipeline leak detection device includes: The pressure detection unit is configured to acquire pipeline pressure timing signals to obtain a pipeline pressure signal dataset. The processing unit is configured to perform pipeline leak detection based on the pressure signal dataset, applying the pipeline leak detection method as described in any one of claims 1 to 9, and output a conclusion on whether the pipeline is leaking.

Images