Energy Transmission Monitoring System and Method Based on Multi-Acoustic Wave Sensors and Sparse Representation

Abstract

The present invention provides an energy transmission monitoring system and method based on multi-acoustic wave sensors and sparse representation, which relates to the technical field of energy transmission. The system includes multi-acoustic wave sensor groups, transmitter modules, network communication modules, data collectors, GPS time calibration modules, and upper computers installed at the first and last monitoring stations of the energy transmission system; the method is based on multi-acoustic wave sensors to perform signal enhancement and working condition error correction, and through sparse representation, jointly reconstructs and solves test samples based on training samples. At the same time, the acoustic wave velocity is updated based on real-time data of multiple sensors, which can well judge the current operating state of the energy transmission system and locate it, shielding working condition adjustment or other abnormal disturbances in pipe sections, so as to improve the detection accuracy and positioning accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of energy transmission, and particularly to an energy transmission monitoring system and method based on multi-acoustic wave sensors and sparse representation. Background Art

[0002] In recent years, with the rapid improvement of the national economy and people's living standards, the demand for energy extraction and transportation has been increasing day by day, and energy transmission systems have been constantly surging. However, many long-distance energy transmission systems still do not have an abnormal detection system or have poor system detection capabilities. Once an abnormal accident occurs, it is extremely easy to cause energy waste, environmental pollution, property losses, and even threaten the lives of the people. Therefore, the abnormal detection technology of energy transmission systems is one of the key technologies to ensure the efficient and stable operation of energy transportation.

[0003] Using the acoustic wave method to detect abnormalities in energy transmission systems is more suitable for long-distance detection compared to current methods such as negative pressure wave detection, etc., and has great engineering application prospects. It is also a hot topic in the current research on abnormal detection of energy transmission systems. However, the signal attenuation and background noise in the process of acoustic wave propagation have always led to a relatively high false alarm rate and missed alarm rate in abnormal judgment. Therefore, it is necessary to innovate the related technologies of the acoustic wave method.

[0004] In the technical solution of Chinese Patent Application No. CN201711113746.3, a gas pipeline abnormal positioning method based on co-located dual sensors is proposed. An intrusive infrasonic wave sensor and a non-intrusive infrasonic wave sensor are installed at the same position at one end of the pipeline to be measured, and an abnormal positioning formula based on co-located dual sensors in the pipeline to be measured is established. The first acoustic wave signal propagating along the gas medium in the pipe is collected by the intrusive infrasonic wave sensor, and the second acoustic wave signal propagating along the pipe wall is collected by the non-intrusive sensor. The time difference between the two infrasonic wave signals is obtained through a timer. If an abnormality occurs, the pipeline abnormality is located using the abnormal positioning formula based on the propagation speeds and time difference of the two infrasonic wave signals. This solves the problem of increased sampling point density caused by the necessity of arranging sensors at both ends of the pipeline.

[0005] The technical solution of Chinese Patent Application No. CN201810862962.6 proposes a pipeline abnormality recognition method, system, and system based on qualitative mapping. An acoustic image map distributed along the pipeline to be measured is calculated and generated based on the infrasonic wave sensing signal and the generation time of the infrasonic wave sensing signal. It is judged whether the attribute value of the acoustic image map is outside the preset abnormal interval constructed based on historical data. When a preset number of attribute values are outside the preset abnormal interval, it is determined that the pipeline to be measured has an abnormality. It can take into account the influence of body noise and external factors on the infrasonic wave detection technology for recognition, thereby improving the accuracy of pipeline abnormality detection and reducing false alarms and missed alarms.

[0006] The technical solution of Chinese invention patent application No. CN202010588123.7 proposes a pipeline condition recognition method for extracting pipeline acoustic signal features based on LCD. The acoustic wave sensor is used to collect pipeline acoustic signals of natural gas pipelines under different conditions respectively, and the locally characteristic-scale decomposition (LCD) method is used to adaptively decompose the collected pipeline acoustic signals. The acoustic signals are decomposed into several intrinsic scale components (ISC) with different scales. The correlation coefficients between each ISC component and the original signal are calculated, and the feature components containing more effective information are selected by using the weighted result of the correlation coefficients. The exponential entropy (EE) of each feature component is calculated to form a feature vector, and then the feature vector is recognized by a pattern recognition method to recognize different conditions of the pipeline.

[0007] The technical solution of Chinese invention application No. CN202010509296.5 proposes an oil and gas pipeline anomaly detection method based on improved VMD and 1DCNN. First, the acoustic wave sensor is used to collect the acoustic signals in natural gas pipelines under normal and abnormal conditions. Then, the parameters of the variational mode decomposition algorithm VMD are initialized, the number of VMD modes K is determined, the acoustic signals are adaptively decomposed, and the IMF with the largest variance is selected as the effective intrinsic mode function. Then, a one-dimensional convolutional neural network 1DCNN pipeline anomaly detection model is constructed by optimizing the network structure and hyperparameters, and the training samples are input into the 1DCNN pipeline anomaly detection model for training. The generalization performance of the model is tested using the test samples, and finally the classification result is obtained to detect whether the pipeline is abnormal. It effectively solves the problem that the decomposition number K of the VMD method depends on empirical selection when decomposing actual signals, and can accurately detect pipeline anomalies and give alarms in time. However, in the above technical solutions, due to signal attenuation and noise interference during the propagation of acoustic waves, it is very difficult to collect acoustic signals and identify abnormal features. On the one hand, long-distance energy transmission will cause the acoustic signals to continuously attenuate until they disappear, and it is necessary to enhance the acoustic signals. On the other hand, since the station operation conditions will interfere with the acoustic signals, it is necessary to identify and correct the condition interference. Therefore, it is necessary to improve the accuracy of acoustic signal collection and the performance of the detection algorithm. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide an energy transmission monitoring system and method based on multi-acoustic wave sensors and sparse representation in view of the above-mentioned deficiencies of the prior art, to solve the problems of long-distance attenuation of acoustic waves and condition interference, and at the same time use sparse representation to fuse and judge the multi-scale features of acoustic signals, and solve the problems of redundancy and difficulty in identifying multi-features of acoustic waves.

[0009] To solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0010] On the one hand, the present invention provides an energy transmission monitoring system based on multi-acoustic wave sensors and sparse representation, including multi-acoustic wave sensor groups, transmitter modules, network communication modules, data collectors, GPS time calibration modules and upper computers installed at the first and last monitoring stations of the energy transmission system;

[0011] The multi-acoustic wave sensor groups at the first monitoring station and the last monitoring station include main sensors and multiple auxiliary sensors;

[0012] The main sensors at the first monitoring station and the last monitoring station are respectively the 2 closest acoustic wave sensors between the first and last monitoring stations;

[0013] The number of the auxiliary sensors has an increasing gradient relationship with the length of the energy transmission pipeline;

[0014] The increasing gradient relationship is that when the length L of the energy transmission pipeline is less than or equal to the set length L0, the number of auxiliary sensors at the first and last monitoring stations is 1; when the pipeline distance L is greater than L0, for every increase of △L in the pipeline length, the number of each group of auxiliary sensors increases by one, that is, the number of auxiliary sensors at the first monitoring station or the last monitoring station in one group is The distance between each sensor at the first monitoring station is When arranging the sensors, the distance between any two sensors is not less than the length d0 corresponding to the detectable time delay of two acoustic wave signals; similarly, the distance between each sensor at the last monitoring station is And

[0015] The transmitter module converts the acoustic signals collected by the acoustic wave sensors, and the data collector is installed in the monitoring station computer room to collect and send the acoustic wave signals received by the transmitter module;

[0016] One set of the transmitter module, data collector, network communication module, GPS time calibration module and upper computer is provided at each of the first and last monitoring stations of the energy transmission system. The acoustic wave sensors are all connected to the transmitter module, data collector and GPS time calibration module at the corresponding end. The GPS time calibration module synchronizes the time of the acoustic wave sensor signals collected by the data collector. The data collector is connected to the upper computer through Ethernet and sends the data with time tags to the upper computer. The upper computers at the first and last monitoring stations communicate through the network communication module;

[0017] The upper computer analyzes and processes the data collected by the data collector to realize the abnormal monitoring of the energy transmission medium;

[0018] The upper computer respectively performs time registration on the acoustic wave signals collected by multiple sensors at the first and last monitoring stations and then performs weighted fusion to obtain the enhanced acoustic wave signals at the first and last monitoring stations; then, for the fusion sample data of the three categories of normal, abnormal and working conditions, the global features and local features of the training samples and the samples to be measured are respectively extracted;

[0019] Based on the features of the training samples and the features of the samples to be measured, joint sparse representation and reconstruction are performed, and the accelerated proximal gradient method is used to alternately update and optimize to obtain the estimated value of the sparse coefficient matrix. For the acoustic wave signals collected and fused in real time by the energy transmission system, a classifier is designed according to the minimum reconstruction error criterion to classify the energy transmission system as abnormal, normal, and operating conditions. Finally, secondary error correction and abnormal positioning of the operating conditions are performed based on the data collected by multiple sensors.

[0020] On the other hand, the present invention also provides an energy transmission monitoring method based on multiple acoustic wave sensors and sparse representation, which specifically includes the following steps:

[0021] Step 1: Multisensor groups are respectively set at the first and last monitoring stations, and the acoustic wave signals X s and Y s collected by the multisensor groups at the first and last monitoring stations are obtained;

[0022] Set the distance between each auxiliary sensor and the main sensor at the first monitoring station as The distance between the main sensor and the last auxiliary sensor is where is the distance between the i-th auxiliary sensor and the previous auxiliary sensor. The acoustic wave signal X s collected by the multisensor group at the last monitoring station is n =[x0, x1, …, x n , where x0 is the acoustic wave signal collected by the main sensor at the first monitoring station, and x1, …, x n are the acoustic wave signals collected by the auxiliary sensors. The number of auxiliary sensors n is related to the length of the energy transmission pipeline; correspondingly, the distance between each auxiliary sensor and the main sensor at the last monitoring station is The distance between the main sensor and the last auxiliary sensor is is the distance between the i-th auxiliary sensor and the previous auxiliary sensor. The acoustic wave signal collected by the multisensor group at the last monitoring station is Y s =[y0, y1, …, y n , where y0 is the acoustic wave signal collected by the main sensor at the last monitoring station, and y1, …, y n are the acoustic wave signals collected by the auxiliary sensors;

[0023] Step 2: Based on the data fusion algorithm with distance weighting of each sensor, the acoustic wave signals X s and Y s collected by the multisensor groups at the first and last monitoring stations are respectively fused to achieve signal enhancement;

[0024] Step 2.1: Use the generalized cross-correlation time-delay estimation to respectively obtain the time delay of each auxiliary sensor relative to the main sensor;

[0025] Set the time delay of each auxiliary sensor relative to the main sensor in the multi-sensor group of the first monitoring station as τ X = [τ x1 , …, τ xn . At the same time, taking the current time of the main sensor as the reference, obtain the time when each auxiliary sensor receives the acoustic signal as t X = [t x1 , …, t xn ; The estimated value of the cross-correlation time delay τ xi between the main sensor and the i-th auxiliary sensor is calculated by the following formula:

[0026]

[0027] where is the cross-correlation coefficient, when obtaining the maximum value, obtain each time delay τ xi , is the weighting function, is the cross-power density spectrum function of the acoustic signals x0 and x i , ω represents the frequency, t represents the time, i = 1, 2, …, n;

[0028] Set the time delay of each auxiliary sensor relative to the main sensor in the multi-sensor group of the last monitoring station as τ Y = [τ y1 , …, τ yn . At the same time, taking the current time of the main sensor as the reference, obtain the time when each auxiliary sensor in the last monitoring station receives the acoustic signal as t Y = [t y1 , …, t yn . The method for solving the time delay of each auxiliary sensor relative to the main sensor is the same as that in the multi-sensor group of the first monitoring station;

[0029] Step 2.2: Determine the weights for fusing the acoustic signals collected by each auxiliary sensor at the first and last monitoring stations based on the distances between the sensors;

[0030] The weights corresponding to the acoustic signals collected by each auxiliary sensor at the first monitoring station are where is the weight corresponding to the i-th auxiliary sensor at the first monitoring station, and

[0031] The weights corresponding to the acoustic signals collected by each auxiliary sensor at the last monitoring station are is the weight corresponding to the i-th auxiliary sensor at the last monitoring station, and

[0032] Step 2.3: Perform weighted fusion on the acoustic signals collected by the multi-sensor groups at the first and last monitoring stations respectively;

[0033] At time t, the signal x(t) after fusing the acoustic wave signals collected by the multi-sensor group of the first monitoring station is shown by the following formula:

[0034]

[0035] Where, is the acoustic wave signal collected by the i-th auxiliary sensor of the first monitoring station at time t shifted forward by time instants;

[0036] At time t, the signal y(t) after fusing the acoustic wave signals collected by the multi-sensor group of the last monitoring station is shown by the following formula:

[0037]

[0038] Where, is the acoustic wave signal collected by the i-th auxiliary sensor of the last monitoring station at time t shifted forward by time instants;

[0039] Step 3: Divide the data of the signals after weighted fusion of the acoustic wave signals collected by the multi-sensor groups of the first and last monitoring stations into a training sample set F1 and a test sample set F2;

[0040] The training sample set F1 consists of the fused acoustic wave data containing three types of labels: normal, abnormal, and working condition; the test sample set F2 includes multiple acoustic wave signal samples containing three types of labels: normal, abnormal, and working condition, and multiple unlabeled fused acoustic wave data samples generated within the set time of the energy transmission pipeline;

[0041] Set the mode categories of the energy transmission system operation as s = [1, 2, 3]. The total numbers p and q of the samples in the training sample set F1 and the test sample set F2 are respectively:

[0042]

[0043] Where, s = [1, 2, 3] respectively represent the three states of normal, abnormal, and working condition of the energy transmission system operation, p I and q J represent the number of samples of the I-th or J-th mode category, q0 is the number of unlabeled samples generated within the set time of the pipeline, and includes the samples generated at the current moment of the system;

[0044] Step 4: Perform multi-scale feature extraction on the training sample set F1 and the test sample set F2 respectively to form a feature vector matrix. The extracted features include global features in the time domain and frequency domain, and local features of time-frequency joint analysis;

[0045] Step 4.1: Determine the global feature vector matrix of the fused acoustic signals in the training sample set F1 and the test sample set F2; this global feature vector matrix is composed of five feature parameters: low-frequency energy ratio, margin factor, sample entropy, mean difference, and spectral kurtosis;

[0046] Step 4.1.1: Obtain the global feature vector matrix corresponding to the training sample set F1 where is the global feature of all samples in the training sample set under the I-th operating state of the energy transmission system, is the global feature vector of a sample in the training sample set, which is obtained from five feature parameters: low-frequency energy ratio, margin factor, sample entropy, mean difference, and spectral kurtosis of the acoustic signal;

[0047] Step 4.1.2: Obtain the global feature vector matrix corresponding to the test sample set F2 as where, is the global feature of all samples in the test sample set under the I-th operating state of the energy transmission system, is the global feature vector of a sample in the test sample set, is the global feature of the unlabeled sample;

[0048] Step 4.2: Decompose the fused acoustic signals in the training sample set F1 and the test sample set F2, screen the components with a correlation greater than the set threshold for demodulation to obtain the envelope spectrum, and finally use the matrix composed of each envelope spectrum as the local feature of the time-frequency joint analysis of the training sample set and the test sample;

[0049] Step 4.2.1: Perform empirical mode decomposition on the fused acoustic signal to obtain N IMF components and a residual component;

[0050] Step 4.2.2: Calculate the cross-correlation coefficient between each IMF component and the original acoustic signal and sort them, and screen the top M IMF components with a correlation greater than the set threshold, M < N;

[0051] Step 4.2.3: For each of the selected IMF components, perform Hilbert transform to obtain the analytic signal z m (t), as shown in the following formula:

[0052] z m (t) = c m (t) + jH[c m (t)]

[0053] where, c m (t) is the selected m-th IMF component, H[] represents the Hilbert transform function, j represents the complex number, and m = 1, 2,..., M;

[0054] The instantaneous amplitude of the analytic signal is obtained, and the envelope signal a(t) of each IMF component is obtained. Then, we have

[0055]

[0056] Finally, the Fourier transform is performed on the envelope signals of each IMF component to obtain the m-th eigenvector on the amplitude spectrum that contains different characteristic frequencies For all samples, we have It is the envelope spectrum of the m-th IMF component of all training samples under the operating state of the I-th energy transmission system; similarly, we obtain It is the envelope spectrum of the m-th IMF component of all test samples under the operating state of the I-th energy transmission system, It is the envelope spectrum of the m-th IMF component of the unlabeled samples generated within q0 set times, where It is the envelope spectrum eigenvector of the m-th IMF component of one of the samples;

[0057] Step 4.2.4: Use the envelope spectra of each IMF component of all samples to form a matrix as the local feature X of the time-frequency joint analysis of the training sample set 1 =[x 1 ,…,x m ,…,x M ; similarly, the local feature Y of the time-frequency joint analysis of the test sample set is obtained 1 =[y 1 ,…,y m ,…,y M ;

[0058] Step 4.3: Jointly combine the global features and local features of the training sample set and the test sample set to obtain the eigenvector matrix X = [X 0 ,X 1 of the training sample set F1 and the eigenvector matrix Y = [Y 0 ,Y 1 of the test sample set F2. The eigenvector matrices X and Y each contain M + 1 characteristic scales;

[0059] Step 5: Use the eigenvector matrix X of the training sample set and the eigenvector matrix Y of the test sample set to perform joint sparse representation and reconstruction on the fused acoustic wave signal;

[0060] Step 5.1: Combine the acoustic wave signal features extracted from the test sample set and the training sample set, and establish a multi-feature learning model with the same structure among multiple features based on the l 1,2 -norm to achieve the joint sparse reconstruction of multiple features. Considering the influence of noise generation, the corresponding optimization problem of the sparse coefficient matrix is:

[0061]

[0062] Then the sparse matrix is represented as W = [w 0 ,..., w k ,..., w M , where represents the sparse coefficient corresponding to the k-th feature, is the sample corresponding to the k-th feature of all target categories in the training sample set, and y k is the sample corresponding to the k-th feature of all target categories in the test sample set. λ is a regularization parameter used to balance the noise and the sparsity of the sparse matrix;

[0063] Step 5.2: Use the accelerated proximal gradient method with the l 1,2 mixed norm to optimize the corresponding optimization problem of the sparse coefficient matrix, and obtain the optimal estimated value of the sparse matrix The specific steps are as follows:

[0064] Step 5.2.1: Input the feature vectors x k ∈ X of multiple training samples, the feature vector y k ∈ Y of the test sample, the regularization parameter λ, and the step size value γ = 2λ, where k = 0,..., M, s = 1, 2, 3;

[0065] Step 5.2.2: Initialize the optimization parameters; Initialize and Set the parameter α0 = 1 and the number of iterations as the sparse coefficient vector corresponding to the k-th feature, as the iteration vector;

[0066] Step 5.2.3: Update the sparse coefficient vector, that is

[0067]

[0068] Step 5.2.4: Update the iteration vector and let That is

[0069]

[0070] Step 5.2.5: Let the number of iterations be incremented by 1, and repeat steps 5.2.3 and 5.2.4 until all iteration vectors and sparse coefficient vectors are updated, and obtain the optimal estimated value of the sparse coefficient matrix and the iteration matrix Specifically as follows:

[0071]

[0072] Step 5.3: According to the training set sample feature vector x k ∈X and the optimal estimated value of the sparse coefficient matrix reconstruct the feature vector of the test sample wherein, represents only retaining the sparse coefficient vector corresponding to the values belonging to the target of the I-th class, and other coefficient elements are 0;

[0073] Step 6: Design a classifier according to the minimum error criterion to identify the states of the unlabeled samples of q0 energy delivery systems including the test sample at the current moment, that is, calculate the reconstruction errors when the test sample is divided into different target categories respectively, and take the target category with the minimum total reconstruction error as the category to which this test sample belongs;

[0074] Step 6.1: Calculate the reconstruction error when the sample to be tested of the energy delivery system and including the sample generated at the current moment is divided into the operating state of the s-th type of energy delivery system as shown in the following formula:

[0075]

[0076] Step 6.2: According to the sparse reconstruction error and the minimum error criterion, the operating state with the minimum total reconstruction error is the state to which the sample to be tested belongs, as shown in the following formula:

[0077]

[0078] wherein, is the state to which the sample to be tested belongs;

[0079] Step 7: Judge the operating state of the energy delivery system at the current moment and correct the working conditions according to the direction of arrival;

[0080] Step 7.1: First judge the category s to which the sample generated at the current moment belongs when the reconstruction error is the smallest. If s = 1, the system is operating normally. If s = 2, the system has an abnormality. If s = 3, the system is in the working condition adjustment state;

[0081] Step 7.2: If the system has an abnormality, perform secondary error correction on the working conditions through the data collected by multiple acoustic sensors. According to the moments t X =[t x1 ,…,t xn and t Y =[t y1 ,…,t yn , judge the direction of the sound source; if the t x and t y moments when the main sensors of the first and last monitoring stations receive the acoustic signals satisfy t x <tx1 <…<t xn and t y <t y1 <…<t yn , it is determined that the system is abnormal; otherwise, the first and last monitoring stations are adjusting the working conditions or other pipe sections are operating abnormally;

[0082] Step 8: Obtain the sound wave propagation speed when the energy transmission system is abnormal, and use cross-correlation time delay estimation to locate the abnormal position;

[0083] Step 8.1: Respectively obtain the propagation speeds of sound waves in the forward and reverse flows in the energy transmission system;

[0084] The distance between each auxiliary sensor and the main sensor at the first monitoring station is The distance between the main sensor and the last auxiliary sensor is The distance between each auxiliary sensor and the main sensor at the last monitoring station is The distance between the main sensor and the last auxiliary sensor is According to the time delays τ X =[τ x1 ,…,τ xn between the main sensor and each auxiliary sensor at the first monitoring station and the time delays τ Y =[τ y1 ,…,τ yn between the main sensor and each auxiliary sensor at the last monitoring station, the reverse flow velocity v X of the sound wave is:

[0085]

[0086] The forward flow velocity v Y of the sound wave is:

[0087]

[0088] Step 8.2: Use the two main sensors at the first and last monitoring stations to find the time delay as the time difference △t for the anomaly to reach the first and last monitoring stations. The time shift corresponding to the maximum value obtained through generalized cross-correlation time delay estimation is the time delay △t for each at the last monitoring station;

[0089] Step 8.3: Set the distance between the first and last monitoring stations as L, then the abnormal position L x is as shown in the following formula:

[0090]

[0091] The beneficial effects of adopting the above technical solutions are as follows: The energy transmission monitoring system and method based on multi-acoustic sensors and sparse representation provided by the present invention perform signal enhancement and working condition error correction based on multi-acoustic sensors, and perform joint reconstruction and solution of test samples based on training samples through sparse representation. At the same time, the acoustic wave velocity is updated based on real-time data of multi-sensors, which can well judge the current operating state of the energy transmission system and locate it, shielding working condition adjustment or other abnormal interferences in pipeline segments, so as to improve the detection accuracy and positioning accuracy. Brief Description of the Drawings

[0092] Figure 1 It is a schematic diagram of the installation of the energy transmission monitoring system device based on multi-acoustic sensors and sparse representation provided by an embodiment of the present invention;

[0093] Figure 2 It is a flowchart of the energy transmission monitoring method based on multi-acoustic sensors and sparse representation provided by an embodiment of the present invention;

[0094] Figure 3 It is a waveform diagram of a group of original acoustic signals collected by a multi-acoustic sensor group in anomaly detection provided by an embodiment of the present invention;

[0095] Figure 4 It is a waveform diagram of a group of acoustic signals after time registration in anomaly detection provided by an embodiment of the present invention;

[0096] Figure 5 It is a flowchart of local feature envelope spectrum extraction provided by an embodiment of the present invention;

[0097] Figure 6 It is a flowchart of joint sparse representation and classification provided by an embodiment of the present invention;

[0098] Figure 7 It is a waveform diagram of enhanced normal and abnormal acoustic signals in anomaly detection provided by an embodiment of the present invention, where (a) is the enhanced normal acoustic signal and (b) is the enhanced abnormal acoustic signal;

[0099] Figure 8 It is a waveform diagram of a certain modal component of the enhanced acoustic signal in anomaly detection provided by an embodiment of the present invention, where (a) is the modal component of the normal acoustic signal and (b) is the modal component of the abnormal acoustic signal;

[0100] Figure 9 It is an envelope spectrum diagram of a certain component of the enhanced acoustic signal in anomaly detection provided by an embodiment of the present invention, where (a) is the envelope spectrum of the normal acoustic signal and (b) is the envelope spectrum of the abnormal acoustic signal.

[0101] In the figure: 1. The main sensor of the first monitoring station; 2. The main sensor of the last monitoring station; 3. The auxiliary sensor of the last monitoring station; 4. The auxiliary sensor of the first monitoring station; 5. The data collector of the first monitoring station; 6. The data collector of the last monitoring station; 7. The GPS time calibration module of the last monitoring station; 8. The GPS time calibration module of the first monitoring station; 9. The network communication module; 10. The upper computer of the first monitoring station; 11. The upper computer of the last monitoring station. Specific implementation manner

[0102] The following combines the accompanying drawings and embodiments to further describe in detail the specific implementation manner of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0103] In this embodiment, an energy transmission monitoring system based on multi-acoustic wave sensors and sparse representation includes multi-acoustic wave sensor groups, transmitter modules, network communication modules, data collectors, GPS time calibration modules, and upper computers installed at the first and last monitoring stations of the energy transmission system;

[0104] The multi-acoustic wave sensor groups of the first monitoring station and the last monitoring station include main sensors and multiple auxiliary sensors;

[0105] The main sensors of the first monitoring station and the last monitoring station are respectively the two nearest acoustic wave sensors between the first and last monitoring stations;

[0106] The number of auxiliary sensors has an increasing gradient relationship with the length of the energy transmission pipeline;

[0107] The increasing gradient relationship is that when the length L of the energy transmission pipeline is less than or equal to the set length L0 (usually 50 km) (the critical value of acoustic wave attenuation is not considered within this set length), the number of auxiliary sensors at the first and last monitoring stations is 1; when the pipeline distance L is greater than L0, for every increase of △L in the pipeline length, the number of auxiliary sensors in each group increases by one, that is, the number of auxiliary sensors at the first monitoring station or the last monitoring station in one group is The distance between each sensor at the first monitoring station is And When arranging the sensors, the distance between any two sensors is not less than the length d0 corresponding to the detectable time delay of the two acoustic wave signals; similarly, the distance between each sensor at the last monitoring station is And The installation of the multi-acoustic wave sensor groups at the first and last monitoring stations is as Figure 1 shown. The first monitoring station includes a main sensor M0 ( Figure 1 shown as label 1 in n ) and n auxiliary sensors M1 - M Figure 1 shown as label 4 in Figure 1 . The multi-acoustic wave sensor group of the last monitoring station includes a main sensor N0 ( n ) and n auxiliary sensors N1 - NFigure 1 as shown by reference numeral 3); in addition, other system devices include data collectors 5 and 6, GPS time calibration modules 7 and 8, upper computers 10 and 11, and network communication module 9. In this embodiment, the number of auxiliary sensors at the first and last monitoring stations is three each, namely auxiliary sensor 1, auxiliary sensor 2, and auxiliary sensor 3.

[0108] The transmitter module converts the acoustic signals collected by the acoustic wave sensors. The data collectors are installed in the monitoring station computer rooms to collect and transmit the acoustic wave signals received by the transmitter module;

[0109] One set of the transmitter module, data collector, network communication module, GPS time calibration module, and upper computer is provided at each of the first and last monitoring stations of the energy transmission system. The acoustic wave sensors are all connected to the transmitter module, data collector, and GPS time calibration module at the corresponding end. The GPS time calibration module synchronizes the time of the acoustic wave sensor signals collected by the data collector. The data collector is connected to the upper computer through Ethernet and sends data with time tags to the upper computer. The upper computers at the first and last monitoring stations communicate with each other through the network communication module;

[0110] The upper computer analyzes and processes the data collected by the data collector to realize the abnormal monitoring of the energy transmission medium;

[0111] The upper computer respectively performs time registration on the acoustic wave signals collected by multiple sensors at the first and last monitoring stations and then performs weighted fusion to obtain the enhanced acoustic wave signals at the first and last monitoring stations; then, for the fusion sample data of the three categories of normal, abnormal, and working conditions, the global features and local features of the training samples and the samples to be measured are respectively extracted; then, based on the features of the training samples and the features of the samples to be measured, joint sparse representation and reconstruction are performed, and the accelerated proximal gradient method is used to alternately update and optimize to obtain the estimated value of the sparse coefficient matrix. For the acoustic wave signals collected and fused in real time by the energy transmission system, a classifier is designed according to the minimum reconstruction error criterion to classify the energy transmission system as abnormal, normal, and working conditions. Finally, secondary error correction and abnormal positioning of the working conditions are performed based on the data collected by multiple sensors.

[0112] In this embodiment, the acoustic wave sensor uses the American PCB106B acoustic wave sensor, and its technical indicators are as follows:

[0113] Range: ±57.2 kPa;

[0114] Sensitivity: 5 mV / kPa;

[0115] Large dynamic step pressure: 1379 kPa;

[0116] Large static pressure: 13790 kPa;

[0117] Resolution: 0.69 Pa;

[0118] Resonant frequency: ≥60 kHz;

[0119] Low - frequency response (-5%): 0.5 Hz;

[0120] Acceleration sensitivity: ≤0.0014 psi / (m / s 2 );

[0121] Temperature range: -54 to 121 °C;

[0122] The data collector is installed in the station yard computer room, and the data collector collects and transmits the signals provided by the transmitter module.

[0123] In this embodiment, the technical index requirements for the data collector are as follows:

[0124] Sampling frequency range: 0 - 2 kHz;

[0125] Number of channels of sampling signals: 4 - 32 channels;

[0126] Sampling accuracy: ≥0.001;

[0127] Communication method: Ethernet;

[0128] Power supply method: 24V DC.

[0129] In this embodiment, the sampling frequency of the acoustic wave signal by each data collector is 1 kHz.

[0130] In this embodiment, the energy transmission monitoring method based on multi - acoustic wave sensors and sparse representation, as Figure 2 shown, specifically includes the following steps:

[0131] Step 1: Set multi - sensor groups at the head and end monitoring stations respectively, and obtain the acoustic wave signals X s and Y s ;

[0132] Set the distance between each auxiliary sensor and the main sensor at the head monitoring station as The distance between the main sensor and the last auxiliary sensor is where, is the distance between the i - th auxiliary sensor and the previous auxiliary sensor, and the acoustic wave signal X s collected by the multi - sensor group at the end monitoring station = [x0, x1, …, x n , x0 is the acoustic wave signal collected by the main sensor at the head monitoring station, x1, …, x n are the acoustic wave signals collected by the auxiliary sensors, and the number of auxiliary sensors n is related to the length of the energy transmission pipeline; similarly, for the end monitoring station, the distance between each auxiliary sensor and the main sensor is The distance between the main sensor and the last auxiliary sensor is is the distance between the i-th auxiliary sensor and the previous auxiliary sensor, and the acoustic wave signals collected by the multi-sensor group at the last monitoring station are Y s =[y0, y1, …, y n , y0 is the acoustic wave signal collected by the main sensor at the last monitoring station, and y1, …, y n are the acoustic wave signals collected by the auxiliary sensors;

[0133] In this embodiment, a set of original acoustic wave signal waveforms collected by the multi-acoustic wave sensor group at the first monitoring station are as Figure 3 shown, and the waveforms after time registration are as Figure 4 shown.

[0134] Step 2: Based on the data fusion algorithm with distance weighting for each sensor, fuse the acoustic wave signals X s and Y s collected by the multi-sensor groups at the first and last monitoring stations respectively to achieve signal enhancement;

[0135] Step 2.1: Use the generalized cross-correlation time delay estimation to obtain the time delay of each auxiliary sensor relative to the main sensor respectively;

[0136] Set the time delay of each auxiliary sensor relative to the main sensor in the multi-sensor group at the first monitoring station as τ X =[τ x1 , …, τ xn , and at the same time, based on the current time of the main sensor, obtain the moments when each auxiliary sensor receives the acoustic wave signal as t X =[t x1 , …, t xn ; The cross-correlation time delay estimation value τ xi between the main sensor and the i-th auxiliary sensor is calculated by the following formula:

[0137]

[0138] where is the cross-correlation coefficient, obtain each time delay τ xi when taking the maximum value, is the weighting function, is the cross-power density spectrum function of the acoustic wave signals x0 and x i , ω represents the frequency, t represents the time, and i = 1, 2, …, n;

[0139] Set the time delay of each auxiliary sensor relative to the main sensor in the multi-sensor group at the last monitoring station as τ Y =[τ y1 , …, τ yn, and at the same time, taking the current time of the main sensor as the reference, the time when each auxiliary sensor at the end monitoring station receives the acoustic wave signal is obtained as t Y = [t y1 ,…,t yn . The time delay of each auxiliary sensor relative to the main sensor is solved in the same way as that in the multi-sensor group of the first monitoring station;

[0140] Step 2.2: Determine the weights for fusing the acoustic wave signals collected by each auxiliary sensor at the first and last monitoring stations based on the distances between the sensors;

[0141] The corresponding weights of the acoustic wave signals collected by each auxiliary sensor at the first monitoring station are where is the weight corresponding to the i-th auxiliary sensor at the first monitoring station, and

[0142] The corresponding weights of the acoustic wave signals collected by each auxiliary sensor at the end monitoring station are is the weight corresponding to the i-th auxiliary sensor at the end monitoring station, and

[0143] Step 2.3: Perform weighted fusion on the acoustic wave signals collected by the multi-sensor groups at the first and last monitoring stations respectively;

[0144] At time t, the signal x(t) after fusing the acoustic wave signals collected by the multi-sensor group at the first monitoring station is shown by the following formula:

[0145]

[0146] where is the acoustic wave signal collected by the i-th auxiliary sensor at the first monitoring station at time t shifted forward time units;

[0147] The signal y(t) after fusing the acoustic wave signals collected by the multi-sensor group at the end monitoring station is shown by the following formula:

[0148]

[0149] where is the acoustic wave signal collected by the i-th auxiliary sensor at the end monitoring station at time t shifted forward time units;

[0150] Step 3: Divide the data of the signals after weighted fusion of the acoustic wave signals collected by the multi-sensor groups at the first and last monitoring stations into a training sample set F1 and a test sample set F2;

[0151] The training sample set F1 consists of fused acoustic wave data containing three types of labels: normal, abnormal, and working conditions; the test sample set F2 includes multiple acoustic wave signal samples containing three types of labels: normal, abnormal, and working conditions, and multiple unlabeled fused acoustic wave data samples generated within the set time of the energy transmission pipeline.

[0152] Set the mode categories of the energy transmission system operation as s = [1, 2, 3]. The total number of samples p and q in the training sample set F1 and the test sample set F2 are respectively:

[0153]

[0154] Among them, s = [1, 2, 3] respectively represent the three states of normal, abnormal, and working conditions of the energy transmission system operation. p I and q J represent the number of samples in the I-th or J-th mode category, q0 is the number of unlabeled samples generated within the set time of the pipeline, and includes the samples generated at the current moment of the system.

[0155] Step 4: Perform multi-scale feature extraction on the training sample set F1 and the test sample set F2 respectively to form a feature vector matrix. The extracted features include global features in the time domain and frequency domain, and local features of time-frequency joint analysis.

[0156] Step 4.1: Determine the global feature vector matrix of the fused acoustic wave signals in the training sample set F1 and the test sample set F2; this global feature vector matrix is composed of five feature parameters: low-frequency energy ratio, margin factor, sample entropy, mean difference, and spectral kurtosis.

[0157] Step 4.1.1: Obtain the global feature vector matrix corresponding to the training sample set F1 Among them is the global feature of all samples in the I-th operating state of the energy transmission system in the training sample set, is the global feature vector of a sample in the training sample set, which is obtained from five feature parameters: low-frequency energy ratio, margin factor, sample entropy, mean difference, and spectral kurtosis of the acoustic wave signal;

[0158] Step 4.1.2: Obtain the global feature vector matrix corresponding to the test sample set F2 as Among them, is the global feature of all samples in the I-th operating state of the energy transmission system in the test sample set, is the global feature vector of a sample in the test sample set, is the global feature of the unlabeled sample;

[0159] Step 4.2: Decompose the fused acoustic signals in the training sample set F1 and the test sample set F2, screen the components with a correlation greater than the set threshold for demodulation to obtain the envelope spectrum, and finally use the matrix composed of each envelope spectrum as the local feature of the time-frequency joint analysis of the training sample set and the test sample, such as Figure 5 as shown;

[0160] Step 4.2.1: Perform empirical mode decomposition (EMD) on the fused acoustic signal to obtain N IMF components and a residual component;

[0161] Step 4.2.2: Calculate the cross-correlation coefficient between each IMF component and the original acoustic signal and sort them, and screen the first M IMF components with a correlation greater than the set threshold, where M < N;

[0162] Step 4.2.3: For each of the selected IMF components, perform Hilbert transform (Hilbert) to obtain the analytic signal z m (t), as shown in the following formula:

[0163] z m (t) = c m (t) + jH[c m (t)]

[0164] where c m (t) is the m-th IMF component selected, H[] represents the Hilbert transform function, j represents the complex number, and m = 1, 2,..., M;

[0165] Then, obtain the envelope signal a(t) of each IMF component by calculating the instantaneous amplitude of the analytic signal, so there is

[0166]

[0167] Finally, perform Fourier transform on the envelope signal of each IMF component to obtain the m-th eigenvector containing different characteristic frequencies on the amplitude spectrum For all samples, there is which is the envelope spectrum of the m-th IMF component of all training samples under the operating state of the I-th energy transmission system; similarly, obtain which is the envelope spectrum of the m-th IMF component of all test samples under the operating state of the I-th energy transmission system, which is the envelope spectrum of the m-th IMF component of the unlabeled samples generated within q0 set times, where is the envelope spectrum eigenvector of the m-th IMF component of one of the samples;

[0168] Step 4.2.4: Use the matrix composed of the envelope spectra of each IMF component of all samples as the local feature X of the time-frequency joint analysis of the training sample set1 = [x 1 , …, x m , …, x M ; Similarly, the local feature Y of the time-frequency joint analysis of the test sample set is obtained 1 = [y 1 , …, y m , …, y M ;

[0169] Step 4.3: Jointly combine the global features and local features of the training sample set and the test sample set respectively to obtain the feature vector matrix X = [X 0 , X 1 of the training sample set F1 and the feature vector matrix Y = [Y 0 , Y 1 of the test sample set F2. The feature vector matrices X and Y each contain M + 1 feature scales;

[0170] In this embodiment, the enhanced normal and abnormal acoustic signals are as Figure 7 shown, the waveform of a certain modal component of the enhanced acoustic signal is as Figure 8 shown, and the envelope spectrum of a certain component of the enhanced acoustic signal is as Figure 9 shown.

[0171] Step 5: Use the feature vector matrix X of the training sample set and the feature vector matrix Y of the test sample set to perform joint sparse representation and reconstruction on the fused acoustic signal, as Figure 6 shown;

[0172] Step 5.1: Combine the acoustic signal features extracted from the test sample set and the training sample set, and establish a multi-feature learning model with the same structure among multiple features based on the l 1,2 norm, and realize the joint sparse reconstruction of multiple features. Considering the influence of noise generation, the corresponding optimization problem of the sparse coefficient matrix is:

[0173]

[0174] Then the sparse matrix is represented as W = [w 0 ,..., w k ,..., w M , where, represents the sparse coefficient corresponding to the kth feature, is the sample corresponding to the kth feature of all target categories in the training sample set, y k is the sample corresponding to the kth feature of all target categories in the test sample set, and λ is a regularization parameter used to balance the noise and the sparsity of the sparse matrix;

[0175] Step 5.2: Adopt l 1,2The accelerated proximal gradient method of the hybrid paradigm optimizes the optimization problem corresponding to the sparse coefficient matrix. In this embodiment, the optimization method Nesterov acceleration algorithm is selected to solve the sparse coefficient matrix W, and the optimal estimated value of the sparse matrix is obtained. The specific steps are as follows:

[0176] Step 5.2.1: Input the feature vectors x k ∈X of multiple training samples, and the feature vector y k ∈Y of the test sample, the regularization parameter λ and the step size value γ = 2λ, where k = 0,..., M, s = 1, 2, 3;

[0177] Step 5.2.2: Initialize the optimization parameters; Initialize and Set the parameter α0 = 1 and the number of iterations as the sparse coefficient vector corresponding to the k-th feature, as the iteration vector;

[0178] Step 5.2.3: Update the sparse coefficient vector, that is

[0179]

[0180] Step 5.2.4: Update the iteration vector and let That is

[0181]

[0182] Step 5.2.5: Let the number of iterations increase by 1, and repeat steps 5.2.3 and 5.2.4 until all iteration vectors and sparse coefficient vectors are updated, and the optimal estimated value of the sparse coefficient matrix is obtained and the iteration matrix Specifically as follows:

[0183]

[0184] Step 5.3: According to the feature vector x k ∈X of the training set samples and the optimal estimation value of the sparse coefficient matrix reconstruct the feature vector of the test sample where, means only retaining the numerical values corresponding to the sparse coefficient vector belonging to the I-th type of target, and other coefficient elements are 0;

[0185] Step 6: Design a classifier according to the minimum error criterion to identify the states of the unlabeled samples of q0 energy delivery systems including the test sample at the current moment, that is, calculate the reconstruction errors when the test sample is divided into different target categories respectively, and take the target category with the minimum total reconstruction error as the category to which this test sample belongs;

[0186] Step 6.1: Calculate the reconstruction error when the sample to be measured of the energy delivery system and including the sample generated at the current moment is divided into the operating state of the sth energy delivery system. As shown in the following formula:

[0187]

[0188] Step 6.2: According to the sparse reconstruction error and the minimum error criterion, the operating state with the minimum total reconstruction error is the state to which the sample to be measured belongs, as shown in the following formula:

[0189]

[0190] where, is the state to which the sample to be measured belongs;

[0191] Step 7: Judge the operating state of the energy delivery system at the current moment and correct the working conditions according to the direction of arrival.

[0192] Step 7.1: First, judge the category s to which the sample generated at the current moment belongs when the reconstruction error is the smallest. If s = 1, the system is operating normally; if s = 2, the system has an abnormality; if s = 3, the system is in the process of working condition adjustment.

[0193] Step 7.2: If the system has an abnormality, perform secondary correction of the working conditions through the data collected by multiple acoustic sensors. According to the moments t X =[t x1 ,…,t xn and t Y =[t y1 ,…,t yn received by the auxiliary sensors, judge the direction of the sound source; if the t x and t y moments when the main sensors of the first and last monitoring stations receive the acoustic signals satisfy t x <t x1 <…<t xn and t y <t y1 <…<t yn , it is judged that the system is abnormal; otherwise, the first and last monitoring stations are in the process of working condition adjustment or other pipe sections are operating abnormally.

[0194] Step 8: Obtain the acoustic wave propagation speed when the energy transmission system is abnormal, and use cross-correlation time delay estimation to locate the abnormal position;

[0195] Step 8.1: Obtain the propagation speeds of the acoustic wave in the forward and reverse directions in the energy transmission system respectively;

[0196] The distance between each auxiliary sensor and the main sensor at the first monitoring station is The distance between the main sensor and the last auxiliary sensor is The distance between each auxiliary sensor and the main sensor at the last monitoring station is The distance between the main sensor and the last auxiliary sensor is According to the time delays τ X =[τ x1 ,…,τ xn between the main sensor and each auxiliary sensor at the first monitoring station and the time delays τ Y =[τ y1 ,…,τ yn between the main sensor and each auxiliary sensor at the last monitoring station, the reverse speed v X of the acoustic wave is:

[0197]

[0198] Similarly, the forward speed v Y of the acoustic wave is:

[0199]

[0200] Step 8.2: Use the two main sensors at the first and last monitoring stations to obtain the time delay as the time difference △t between the abnormal situation reaching the first and last monitoring stations, and the time shift corresponding to the maximum value obtained through generalized cross-correlation time delay estimation is the time delay △t of each at the last monitoring station;

[0201] Step 8.3: Set the distance between the first and last monitoring stations as L, then the abnormal position (the distance from the first monitoring station) L x is as shown in the following formula:

[0202]

[0203] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope defined by the claims of the present invention.

Claims

1. An energy transmission monitoring system based on multi-acoustic wave sensors and sparse representation, characterized by: It includes a multi - acoustic wave sensor group, a transmitter module, a network communication module, a data collector, a GPS time - calibration module, and a host computer installed at the first monitoring station and the last monitoring station of the energy transmission system; The multi - acoustic wave sensor groups at the first monitoring station and the last monitoring station include a main sensor and multiple auxiliary sensors; The main sensors at the first monitoring station and the last monitoring station are respectively the two closest acoustic wave sensors between the first and the last monitoring stations; The number of the auxiliary sensors has an increasing gradient relationship with the length of the energy transmission pipeline; The transmitter module converts the acoustic signals collected by the acoustic wave sensors. The data collector is installed in the monitoring station computer room to collect and send the acoustic wave signals received by the transmitter module; One set of the transmitter module, the data collector, the network communication module, the GPS time - calibration module, and the host computer is provided at each of the first and the last monitoring stations of the energy transmission system. The acoustic wave sensors are all connected to the transmitter module, the data collector, and the GPS time - calibration module at the corresponding end. The GPS time - calibration module synchronizes the time of the acoustic wave sensor signals collected by the data collector. The data collector is connected to the host computer through Ethernet and sends data with time tags to the host computer. The host computers at the first and the last monitoring stations communicate with each other through the network communication module; The host computer analyzes and processes the data collected by the data collector to realize the abnormal monitoring of the energy transmission medium. The specific method is as follows: The host computer respectively performs time registration on the acoustic wave signals collected by multiple sensors at the first and the last monitoring stations and then performs weighted fusion to obtain the enhanced acoustic wave signals at the first and the last monitoring stations. Then, for the fusion sample data of three categories: normal, abnormal, and working condition, the global features and local features of the training samples and the samples to be measured are respectively extracted; Then, based on the features of the training samples and the samples to be measured, joint sparse representation and reconstruction are carried out. The accelerated proximal gradient method is used to alternately update and optimize to obtain the estimated value of the sparse coefficient matrix. For the acoustic wave signals collected and fused in real - time by the energy transmission system, a classifier is designed according to the minimum reconstruction error criterion to classify the energy transmission system as abnormal, normal, and working condition. Finally, secondary error correction of the working condition and abnormal positioning are carried out based on the data collected by multiple sensors.

2. The energy transmission monitoring system based on multi-acoustic wave sensors and sparse representation according to claim 1, characterized in that: The increasing gradient relationship between the number of the auxiliary sensors and the length of the energy transmission pipeline is as follows: When the length L of the energy transmission pipeline is less than or equal to the set length L0, the number of auxiliary sensors at the first and last monitoring stations is 1; when the pipeline distance L is greater than L0, for every increase of △L in the pipeline length, the number of auxiliary sensors in each group increases by one, that is, the number of auxiliary sensors at the first monitoring station or the last monitoring station in a group is The distance between the sensors at the first monitoring station is and When arranging the sensors, the distance between any two sensors is not less than the length d0 corresponding to the detectable time delay of the two acoustic signals; similarly, the distance between the sensors at the last monitoring station is and 3. A method for monitoring energy transmission based on multi-acoustic wave sensors and sparse representation, implemented based on the system described in claim 1, characterized in that: It includes the following steps: Step 1: Set multi-acoustic wave sensor groups at the head and end monitoring stations respectively, and obtain the acoustic wave signals X s and Y s ; Set the distance between each auxiliary sensor and the main sensor at the first monitoring station to be The distance between the main sensor and the last auxiliary sensor is Among them, is the distance between the i-th auxiliary sensor and the previous auxiliary sensor. The acoustic signal X collected by the multi-acoustic sensor group at the last monitoring station s =[x0, x1, …, x n , where x0 is the acoustic signal collected by the main sensor at the first monitoring station, and x1, …, x n are the acoustic signals collected by the auxiliary sensors. The number n of auxiliary sensors is related to the length of the energy transmission pipeline; the corresponding distance between each auxiliary sensor and the main sensor at the last monitoring station is The distance between the main sensor and the last auxiliary sensor is is the distance between the i-th auxiliary sensor and the previous auxiliary sensor. The acoustic signal collected by the multi-acoustic sensor group at the last monitoring station is Y s =[y0, y1, …, y n , where y0 is the acoustic signal collected by the main sensor at the last monitoring station, and y1, …, y n are the acoustic signals collected by the auxiliary sensors; Step 2: Based on the data fusion algorithm with distance weighting for each sensor, fuse the acoustic signals X s and Y s collected by the multi-acoustic wave sensor groups at the first and last monitoring stations respectively to achieve signal enhancement; Step 3: Divide the data of the signals after weighted fusion of the acoustic wave signals collected by the multi - acoustic wave sensor groups at the first and the last monitoring stations into a training sample set F1 and a test sample set F2; The training sample set F1 consists of the fused acoustic wave data containing three types of labels: normal, abnormal, and working condition. The test sample set F2 includes multiple acoustic wave signal samples containing three types of labels: normal, abnormal, and working condition and multiple unlabeled fused acoustic wave data samples generated within the set time of the energy transmission pipeline; Set the mode categories of the operation of the energy transmission system as s = [1, 2, 3]. The total number of samples p and q of the training sample set F1 and the test sample set F2 are respectively: Among them, s = [1, 2, 3] respectively represent the three states of normal, abnormal, and operating conditions of the energy transmission system operation, p I and q J represent the number of samples of the I or J type of pattern class, q0 is the number of unlabeled samples generated within the pipeline setting time, and includes the samples generated at the current moment of the system; Step 4: Respectively perform multi - scale feature extraction on the training sample set F1 and the test sample set F2 to form a feature vector matrix. The extracted features include global features in the time domain and frequency domain and local features of time - frequency joint analysis; Step 5: Use the feature vector matrix X of the training sample set and the feature vector matrix Y of the test sample set to perform joint sparse representation and reconstruction on the fused acoustic wave signal; Step 6: Design a classifier according to the minimum reconstruction error criterion to identify the states of the q0 unlabeled samples of the energy transmission system including the test sample at the current moment, that is, calculate the reconstruction errors when the test sample is divided into different target categories respectively, and take the target category with the minimum total reconstruction error as the category to which this test sample belongs; Step 7: Judge the operating state of the energy transmission system at the current moment and correct the working conditions according to the direction of arrival of the wave; Step 8: Obtain the acoustic wave propagation speed when the energy transmission system is abnormal, and use cross-correlation time delay estimation to locate the abnormal position.

4. The energy delivery monitoring method based on multi-acoustic wave sensors and sparse representation according to claim 3, characterized in that: The specific method of the said Step 2 is as follows: Step 2.1: Use generalized cross-correlation time delay estimation to obtain the time delay of each auxiliary sensor relative to the main sensor respectively; Set the time delay of each auxiliary sensor relative to the main sensor in the multi-acoustic wave sensor group of the first monitoring station to be τ X = [τ x1 , …, τ xn , and at the same time, taking the current time of the main sensor as the reference, obtain the time when each auxiliary sensor receives the acoustic wave signal as t X = [t x1 , …, t xn ; The estimated value of the cross-correlation time delay τ xi between the main sensor and the i-th auxiliary sensor is calculated by the following formula: Among them, is the cross-correlation coefficient, each time delay τ is obtained when the maximum value is achieved xi , is the weighting function, is the cross-power density spectrum function of the acoustic wave signals x0 and x i , ω represents frequency, t represents time, i = 1, 2, …, n; Set the time delay of each auxiliary sensor relative to the main sensor in the multi-acoustic wave sensor group of the end monitoring station to be τ Y = [τ y1 , …, τ yn . At the same time, taking the current time of the main sensor as the reference, obtain the time when each auxiliary sensor in the end monitoring station receives the acoustic wave signal as t Y = [t y1 , …, t yn . The method for solving the time delay of each auxiliary sensor relative to the main sensor is the same as that in the multi-acoustic wave sensor group of the head monitoring station; Step 2.2: Based on the distances between the sensors, determine the weights for fusing the acoustic wave signals collected by the auxiliary sensors at the head and end monitoring stations; The corresponding weights of the acoustic signals collected by each auxiliary sensor of the first monitoring station are wherein, is the weight corresponding to the i-th auxiliary sensor of the first monitoring station, and The corresponding weights of the acoustic signals collected by each auxiliary sensor of the end monitoring station are is the weight corresponding to the i-th auxiliary sensor of the end monitoring station, and Step 2.3: Perform weighted fusion on the acoustic wave signals collected by the multi-acoustic wave sensor groups at the head and end monitoring stations respectively; At time t, the fused signal x(t) of the acoustic wave signals collected by the multi-acoustic wave sensor group at the head monitoring station is shown in the following formula: Among them, is the acoustic wave signal collected by the i-th auxiliary sensor of the first monitoring station at time t shifted forward by a number of times; The fused signal y(t) of the acoustic wave signals collected by the multi-acoustic wave sensor group at the end monitoring station is shown in the following formula: Among them, is the acoustic wave signal collected by the i-th auxiliary sensor of the monitoring station at the end of time t shifted forward time instants.

5. The energy delivery monitoring method based on multi-acoustic wave sensors and sparse representation according to claim 4, characterized in that: The specific method of the said Step 4 is as follows: Step 4.1: Determine the global feature vector matrix of the fused acoustic wave signals in the training sample set F1 and the test sample set F2; This global feature vector matrix is composed of five feature parameters: low-frequency energy ratio, margin factor, sample entropy, mean difference, and spectral kurtosis; Step 4.1.1: Obtain the global feature vector matrix corresponding to the training sample set F1 where is the global feature of all samples in the I-th operating state of the energy transmission system in the training sample set, is the global feature vector of a sample in the training sample set, which is obtained from five feature parameters: the low-frequency energy ratio of the acoustic signal, the margin factor, the sample entropy, the mean difference, and the spectral kurtosis; Step 4.1.2: Obtain the global feature vector matrix corresponding to the test sample set F2 as where is the global feature of all samples in the I-th operating state of the energy transmission system in the test sample set, is the global feature vector of a sample in the test sample set, is the global feature of the unlabeled sample; Step 4.2: Decompose the fused acoustic wave signals in the training sample set F1 and the test sample set F2, screen the components with a correlation greater than the set threshold for demodulation to obtain the envelope spectrum, and finally use the matrix composed of each envelope spectrum as the local feature of the time-frequency joint analysis of the training sample set and the test sample; Step 4.2.1: Perform empirical mode decomposition on the fused acoustic wave signal to obtain N IMF components and a residual component; Step 4.2.2: Calculate the cross-correlation coefficients of each IMF component with the original acoustic wave signal and sort them, and screen the first M IMF components with a correlation greater than the set threshold, where M < N; Step 4.2.3: For each of the selected IMF components, perform Hilbert transform respectively to obtain the analytic signal z m (t), as shown in the following formula: z m z(t) = c m z(t) + jHz m (t)]; where c m (t) is the m-th IMF component selected, H[] represents the Hilbert transform function, j represents the complex number, and m = 1, 2, …, M; Then, obtain the instantaneous amplitude of the analytic signal to get the envelope signal a(t) of each IMF component, then Finally, perform Fourier transform on the envelope signals of each IMF component to obtain the m-th eigenvector on the amplitude spectrum containing different characteristic frequencies. For all samples, there is which is the envelope spectrum of the m-th IMF component of all training samples under the operating state of the I-th energy transmission system; similarly, obtain which is the envelope spectrum of the m-th IMF component of all test samples under the operating state of the I-th energy transmission system, which is the envelope spectrum of the m-th IMF component of the unlabeled samples generated within q0 set time periods, where is the envelope spectrum eigenvector of the m-th IMF component of one of the samples; Step 4.2.4: When constructing a matrix with the envelope spectra of the IMF components of all samples as the local feature X of the time-frequency joint analysis 1 = [x 1 , …, x m , …, x M ; Similarly, obtain the local feature Y of the time-frequency joint analysis of the test sample set 1 = [y 1 , …, y m , …, y M ; Step 4.3: Jointly process the global features and local features of the training sample set and the test sample set respectively to obtain the feature vector matrix X = [X 0 , X 1 of the training sample set F1 and the feature vector matrix Y = [Y 0 , Y 1 of the test sample set F2. The feature vector matrices X and Y respectively contain M + 1 feature scales.

6. The energy delivery monitoring method based on multi-acoustic wave sensors and sparse representation according to claim 5, characterized in that: The specific method of the said Step 5 is as follows: Step 5.1: Combine the acoustic signal features extracted from the test sample set and the training sample set, and establish a multi-feature learning model with the same structure among multiple features based on the l 1,2 norm, and realize the joint sparse reconstruction of multiple features. Considering the influence of noise generation, the corresponding optimization problem of the sparse coefficient matrix is as follows: Then the sparse matrix is represented as W = [w 0 ,..., w k ,..., w M , where represents the sparse coefficient corresponding to the k-th feature, is the sample corresponding to the k-th feature of all target categories in the training sample set, y k is the sample corresponding to the k-th feature of all target categories in the test sample set, and λ is a regularization parameter used to balance the noise and the sparsity of the sparse matrix; Step 5.2: Use l 1,2 The accelerated proximal gradient method of the hybrid paradigm is used to optimize the corresponding optimization problem of the sparse coefficient matrix to obtain the optimal estimated value of the sparse matrix The specific steps are as follows: Step 5.2.1: Input the feature vectors x of multiple training samples k ∈ X, and the feature vector y of the test sample k ∈ Y, regularization parameter λ and step size value γ = 2λ, where k = 0, …, M, s = 1, 2, 3; Step 5.2.2: Initialize the optimization parameters; Initialize and Set the parameter α0 = 1 and the number of iterations as the sparse coefficient vector corresponding to the k-th feature, as the iteration vector; Step 5.2.3: Update the sparse coefficient vector, that is Step 5.2.4: Update the iteration vector And let That is Step 5.2.5: Let the iteration number be incremented by 1, and repeat Steps 5.2.3 and 5.2.4 until all the iterative vectors and sparse coefficient vectors are updated, obtaining the optimal estimated value of the sparse coefficient matrix and the iterative matrix Specifically as follows: Step 5.3: According to the training set sample feature vector x k ∈X and the optimal estimated value of the sparse coefficient matrix reconstruct the feature vector of the test sample wherein represents only retaining the sparse coefficient vector corresponding to the values belonging to the targets of the I-th class, and other coefficient elements are 0.

7. The energy delivery monitoring method based on multi-acoustic wave sensors and sparse representation according to claim 6, wherein: The specific method of the said Step 6 is as follows: Step 6.1: Calculate the reconstruction error when the sample to be measured in the energy delivery system and including the sample generated at the current moment is classified as the operating state of the s-th type of energy delivery system As shown in the following formula: Step 6.2: According to the sparse reconstruction error and the minimum error criterion, the operating state with the minimum total reconstruction error is the state to which the sample to be tested belongs, as shown in the following formula: Among them, is the state to which the sample to be measured belongs.

8. The energy transmission monitoring method based on multi-acoustic wave sensors and sparse representation according to claim 7, characterized in that: The specific method of the said Step 7 is as follows: Step 7.1: First, judge the category s to which the sample generated at the current moment belongs when the reconstruction error is the smallest. If s = 1, the system is operating normally; if s = 2, the system has an abnormality; if s = 3, the system is in the working condition adjustment state; Step 7.2: If the system encounters an anomaly, perform secondary error correction on the working conditions based on the data collected by the multi-acoustic wave sensors. According to the time t when each auxiliary sensor receives the acoustic wave signal X =[t x1 ,…,t xn and t Y =[t y1 ,…,t yn , determine the direction of the sound source. If the t x and t y when the main sensors of the first and last monitoring stations receive the acoustic wave signal satisfy t x <t x1 <…<t xn and t y <t y1 <…<t yn , then it is determined that the system is abnormal; otherwise, the first and last monitoring stations are performing working condition adjustment or other pipe sections are operating abnormally.

9. The energy delivery monitoring method based on multi-acoustic wave sensors and sparse representation according to claim 8, wherein: The specific method of the said Step 8 is as follows: Step 8.1: Obtain the propagation speeds of the acoustic wave in the forward and reverse flows in the energy transmission system respectively; The distance between each auxiliary sensor and the main sensor at the first station is The distance between the main sensor and the last auxiliary sensor is The distance between each auxiliary sensor and the main sensor at the last station is The distance between the main sensor and the last auxiliary sensor is According to the time delays τ X = [τ x1 , …, τ xn between the main sensor and each auxiliary sensor at the first station and the time delays τ Y = [τ y1 , …, τ yn between the main sensor and each auxiliary sensor at the last station, the upstream velocity v X of the acoustic wave is: The downstream velocity v of the sound wave Y is as follows: Step 8.2: Use the two main sensors of the first and last monitoring stations to calculate the time delay as the time difference △t of the abnormal arrival at the first and last monitoring stations. The time shift corresponding to the maximum value obtained by the generalized cross-correlation time delay estimation is the time delay △t of each of the last monitoring stations; Step 8.3: Set the distance between the first and last monitoring stations as L, then the abnormal position is L x As shown in the following formula:

Images