A method and system for detecting leaks in a water distribution network using optical fiber vibration auscultation

Abstract

The application provides a kind of water supply network leak optical fiber vibration auscultation method and system, it is related to water supply network monitoring technical field, this method includes: using FPGA main clock to double channel optical fiber vibration signal is synchronously sampled, to obtain time alignment pipeline vibration signal and environmental noise signal;Pipeline vibration signal and environmental noise signal are carried out multidimensional feature fusion, to obtain pipeline leakage characteristic vector;Pipeline leakage characteristic vector is input into CNN-LSTM hybrid model, to obtain the pipeline leakage information that CNN-LSTM hybrid model is exported after feature analysis and prediction processing to pipeline leakage characteristic vector.Solved the technical problem that the recognition accuracy of water supply pipeline leakage in complex environment is lower in the related art, reached the technical effect of improving the recognition accuracy of water supply pipeline leakage in complex environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water supply network monitoring, in particular to a fiber vibration auscultation method and system for water supply network leakage. BACKGROUND

[0002] With the acceleration of urbanization, the scale of water supply network continues to expand. As the core artery of urban infrastructure, it bears the heavy responsibility of water resource allocation in the field of residents' life water, industrial production and public services. However, the water supply network faces many severe challenges in long-term operation. A large number of old pipes serving for more than 20 years are prone to corrosion due to material aging, such as rusting of cast iron pipes and brittleness of PE pipes, resulting in a leakage rate of 10%-30%, causing more than 20 billion tons of water resources to be wasted every year.

[0003] In related technologies, water supply network leakage detection mainly relies on three methods. The first method is pressure wave method, which determines leakage by sudden change of pipeline pressure, but is easily disturbed by water flow turbulence, pump valve start-stop and other factors, and cannot realize continuous monitoring in large diameter main pipes above DN800 due to rapid pressure attenuation, with a missed detection rate of more than 40%. The second method is sound listening method, which relies on artificial holding of listening leak detector for night patrol, is disturbed by traffic noise and complex underground pipeline vibration, and has an identification accuracy of less than 30% and a positioning deviation of 5-8 meters. The third method is optical fiber sensing method, which can monitor pipeline vibration, but the signal collected by a single sensing optical fiber has serious overlap between environmental noise spectrum (10-500Hz) and leakage characteristic frequency band (0.1-200Hz), making it difficult to distinguish effective leakage signal from traffic / construction noise, especially in areas with ground subsidence (annual subsidence >3cm), where pipeline deformation leads to a time delay error of sound wave propagation of more than ±15ms, further amplifying the positioning deviation.

[0004] However, using the above water supply network leakage detection methods, due to the lack of targeted environmental noise elimination mechanism and the existing signal processing method cannot effectively separate the overlapping spectrum, resulting in serious interference of effective leakage signal extraction by environmental noise, and further leading to low identification accuracy of water supply pipeline leakage in complex environment in related technologies. SUMMARY

[0005] The present application provides a fiber vibration auscultation method and system for water supply network leakage, which is used to improve the identification accuracy of water supply pipeline leakage in complex environment.

[0006] In a first aspect, the present application provides a fiber vibration auscultation method for water supply network leakage, applied to the fiber vibration auscultation system, the method comprising: synchronously sampling the double-channel fiber vibration signals by using the FPGA main clock to obtain the time-aligned pipeline vibration signals and environmental noise signals; performing multi-dimensional feature fusion on the pipeline vibration signals and the environmental noise signals to obtain a pipeline leakage feature vector; inputting the pipeline leakage feature vector into a CNN-LSTM hybrid model to obtain pipeline leakage information output by the CNN-LSTM hybrid model after performing feature analysis and prediction processing on the pipeline leakage feature vector.

[0007] By using the above technical solution, the double-channel fiber vibration signals are synchronously sampled by using the FPGA main clock to directly obtain the time-aligned pipeline vibration signals and environmental noise signals. The signal misalignment problem caused by different clock synchronization in the traditional sampling mode can be avoided, laying a foundation for subsequent accurate analysis of pipeline vibration characteristics and distinguishing between effective signals and environmental noise, thereby ensuring the effectiveness and reliability of the leakage detection data from the sampling source. Furthermore, the technical problem of low recognition accuracy of water supply pipeline leakage in a complex environment in the related art is solved, and the technical effect of improving the recognition accuracy of water supply pipeline leakage in a complex environment is achieved.

[0008] In a second aspect, the present application provides a fiber vibration auscultation system, comprising: one or more processors and a memory; the memory is coupled with the one or more processors, and the memory is configured to store computer program codes, the computer program codes comprising computer instructions, and the one or more processors invoke the computer instructions to enable the fiber vibration auscultation system to perform the method described in the first aspect and any possible implementation manner of the first aspect.

[0009] In a third aspect, the present application provides a computer program product comprising instructions, when the computer program product is executed on the fiber vibration auscultation system, the fiber vibration auscultation system performs the method described in the first aspect and any possible implementation manner of the first aspect.

[0010] In a fourth aspect, the present application provides a computer-readable storage medium comprising instructions, when the instructions are executed on the fiber vibration auscultation system, the fiber vibration auscultation system performs the method described in the first aspect and any possible implementation manner of the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 FIG. 1 is a flowchart of a fiber vibration auscultation method for water supply network leakage in the present application;

[0012] Figure 2Fig. 1 is a schematic diagram of a layered architecture of a fiber vibration auscultation system according to an embodiment of the present application;

[0013] Figure 3 Fig. 2 is a schematic diagram of a hardware composition of a fiber vibration auscultation system according to an embodiment of the present application;

[0014] Figure 4 Fig. 3 is a schematic diagram of a functional module of a fiber vibration auscultation system according to an embodiment of the present application;

[0015] Figure 5 Fig. 4 is a schematic diagram of a function of an alarm reaching module according to an embodiment of the present application;

[0016] Figure 6 Fig. 5 is a schematic diagram of a function of a diagnosis analysis module according to an embodiment of the present application;

[0017] Figure 7 Fig. 6 is a schematic diagram of a principle of fiber Rayleigh scattering measurement according to an embodiment of the present application;

[0018] Figure 8 Fig. 7 is a schematic diagram of a structure of a CNN-LSTM hybrid model according to an embodiment of the present application;

[0019] Figure 9 Fig. 8 is a schematic diagram of an operation of a convolution kernel according to an embodiment of the present application;

[0020] Figure 10 Fig. 9 is a schematic diagram of a structure of LSTM time series modeling according to an embodiment of the present application;

[0021] Figure 11 Fig. 10 is a schematic diagram of an architecture of a multi-task learning neural network according to an embodiment of the present application;

[0022] Figure 12 Fig. 11 is a schematic diagram of a physical device structure of a fiber vibration auscultation system according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The terminology used in the following embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in the specification and the appended claims, the singular forms "a," "an" and "the" are intended to include both singular and plural forms, unless the context clearly indicates otherwise. It will be further understood that the terms "and / or," as used herein, refers to

[0024] The terms "first", "second", "third", etc. are used only for the purpose of description, and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0025] The present application provides a fiber vibration auscultation method for water supply network leakage, referring to Figure 1 , Figure 1 is a flowchart of the fiber vibration auscultation method for water supply network leakage in the embodiments of the present application, comprising the following steps:

[0026] Step S101, synchronously sampling the double-channel fiber vibration signals by using the FPGA main clock to obtain time-aligned pipeline vibration signals and environmental noise signals;

[0027] Step S102, performing multi-dimensional feature fusion on the pipeline vibration signals and the environmental noise signals to obtain a pipeline leakage feature vector;

[0028] Step S103, inputting the pipeline leakage feature vector into the CNN-LSTM hybrid model to obtain pipeline leakage information output by the CNN-LSTM hybrid model after performing feature analysis and prediction processing on the pipeline leakage feature vector.

[0029] In the above embodiments, the FPGA main clock refers to the main clock circuit in the field programmable gate array, which is used to provide a unified clock reference signal for the fiber vibration auscultation system. The double-channel fiber vibration signal represents the vibration signal collected by two independent fiber channels simultaneously, including the pipeline vibration signal and the environmental noise signal. Synchronous sampling means sampling the signals of the two channels according to the same clock beat. Time alignment means that the signals collected by the two channels have a strict corresponding relationship in the time dimension. Multi-dimensional feature fusion means combining and processing feature information in multiple dimensions. The CNN-LSTM hybrid model represents a composite deep learning model combining convolutional neural networks and long short-term memory networks.

[0030] In the above embodiment, in a large urban water supply system, in order to timely find the leakage of the water supply network, the optical fiber vibration auscultation method is used for monitoring. In the water supply area, key nodes are selected, and double-channel optical fiber sensors are installed on the surface of the water supply pipeline and the environment away from the pipeline. The FPGA main clock samples at a frequency of 100 kHz, synchronously samples the pipeline vibration signals collected by the optical fiber sensor on the pipeline surface and the environmental noise signals collected by the environmental optical fiber sensor, and after A / D conversion, time-aligned digital signal sequences are obtained. For example, on a DN800 water supply pipeline 5 kilometers long, a monitoring point is set every 500 meters, and double-channel signal collection is performed at each monitoring point to ensure that pipeline vibration information can be fully captured. The pipeline vibration signals and environmental noise signals obtained by sampling are input to the feature extraction module. The signal is analyzed in the time domain to extract features such as peak value, mean value, and variance; in the frequency domain, the signal spectrum features such as energy distribution and main frequency are obtained through fast Fourier transform (FFT); at the same time, time-frequency domain analysis is performed, and short-time Fourier transform (STFT) or wavelet transform is used to obtain the characteristics of the signal at different times and frequencies. Fuse these time domain, frequency domain and time-frequency domain features to construct a pipeline leakage feature vector containing 20 dimensions. For example, when analyzing a certain collection signal, it is found that the pipeline vibration signal has an abnormal energy concentration in the low frequency band, while the environmental noise signal has a low energy in this frequency band. Through feature fusion, this difference is clearly reflected in the feature vector. The constructed pipeline leakage feature vector is input into the pre-trained CNN-LSTM hybrid model. The CNN part first performs convolution operation on the feature vector, extracts local features through different size convolution kernels, reduces the data dimension through the pooling layer and retains the key features; then the features extracted by CNN are transmitted to the LSTM layer, and LSTM uses its memory unit to analyze and learn the time sequence information of the features, and predicts whether the pipeline leaks and the location and degree of the leakage.

[0031] In the above embodiments, in the signal acquisition stage, an adaptive filtering algorithm such as the least mean square (LMS) algorithm is introduced. According to the real-time collected environmental noise signal, the filter parameters are dynamically adjusted, and the pipeline vibration signal is denoised. This can better adapt to the complex and changeable environmental noise, further improve the quality of the pipeline vibration signal, and provide more accurate data for subsequent feature extraction and analysis. For example, when there is strong noise interference near the monitoring area due to construction, the adaptive filtering algorithm can quickly adjust and effectively suppress the influence of noise on the pipeline vibration signal. In the multi-dimensional feature fusion process, a multi-scale analysis method is used. In addition to the conventional time domain, frequency domain and time-frequency domain features, the signal is analyzed from different scales, such as using multi-scale decomposition technology to obtain features of the signal at different resolutions. These multi-scale features are fused with the original features to construct a more rich and representative pipeline leakage feature vector. This helps to capture the subtle feature differences of different degrees and types of leakage, and improves the recognition ability of the model for various leakage conditions. For example, for slow leakage of small diameter pipelines and sudden leakage of large diameter pipelines, multi-scale feature fusion can more accurately distinguish their features and improve the accuracy of detection. In view of the changing conditions of pipeline state, environmental conditions and other conditions during the operation of the water supply network, the dynamic updating mechanism of the CNN-LSTM hybrid model is used. New detection data is collected regularly and added to the training set to retrain the model. At the same time, online learning algorithm is used to make the CNN-LSTM hybrid model adjust parameters in real time according to new data and adapt to the changes of pipeline operation state. This can ensure that the model always maintains high detection accuracy and reliability during long-term use, and avoids performance degradation due to changes in the pipeline network. For example, when the vibration characteristics of a section of pipeline change due to aging, the dynamic updating mechanism of the model can learn these changes in time and adjust the detection strategy to ensure the accuracy of leakage detection. The optical fiber vibration auscultation method is integrated with the geographic information system (GIS). After detecting pipeline leakage, the leakage location information output by the CNN-LSTM hybrid model is directly labeled on the GIS map, and the relevant attribute information of the pipeline such as pipe diameter, material, burial depth, etc. is also displayed. The staff can intuitively view the leakage location and surrounding environment through the GIS system, quickly develop a repair plan, and improve the repair efficiency. In addition, the GIS system can also analyze historical leakage data, predict areas where leakage may occur in the future, and take preventive and maintenance measures in advance to further ensure the safe operation of the water supply network.

[0032] Through the above steps, the double-channel optical fiber vibration signals are synchronously sampled by using the FPGA main clock, and the time-aligned pipeline vibration signals and environmental noise signals are directly obtained. The signal misalignment problem caused by different clock in the traditional sampling mode can be avoided, which lays a foundation for subsequent accurate analysis of pipeline vibration characteristics and distinguishes effective signals from environmental noise, thereby ensuring the effectiveness and reliability of the leakage detection data from the sampling source. Further, the technical problem of low recognition accuracy of water supply pipeline leakage in a complex environment in the related art is solved, and the technical effect of improving the recognition accuracy of water supply pipeline leakage in a complex environment is achieved.

[0033] In the above steps, the execution subject can be a system with water supply pipeline leakage detection capability, such as an optical fiber vibration auscultation system, or a device with water supply pipeline leakage detection capability, or a controller or processor in the device or system, or a separate controller or processor, or other processing devices or processing units with similar processing functions, but is not limited thereto.

[0034] In an optional embodiment, the pipeline vibration signals and the environmental noise signals are subjected to multi-dimensional feature fusion to obtain a pipeline leakage feature vector, specifically including: performing optical time domain reflection processing on the pipeline vibration signals to obtain a phase delay variation of backscattered Rayleigh light; generating a local oscillation light by using an ultra-narrow linewidth pulsed laser, and superimposing the backscattered Rayleigh light and the local oscillation light on a photodetector to obtain an interference light intensity signal; performing filtering processing on the interference light intensity signal and the environmental noise signal by using an improved LMS adaptive filtering algorithm to obtain a leakage vibration signal; determining an optical fiber strain feature of a first optical fiber channel in the double-channel optical fiber vibration signals according to the phase delay variation, wherein the first optical fiber channel is tightly coupled with the water supply pipeline and is used for collecting the pipeline vibration signals, and a second optical fiber channel in the double-channel optical fiber vibration signals maintains a preset distance from the water supply pipeline and is used for collecting the environmental noise signals; performing time domain feature extraction on the leakage vibration signal to obtain a leakage impact feature, and performing frequency domain feature extraction on the leakage vibration signal to obtain a leakage voiceprint feature; and performing multi-dimensional feature fusion on the optical fiber strain feature, the leakage impact feature and the leakage voiceprint feature to obtain the pipeline leakage feature vector.

[0035] In the above embodiments, the optical time domain reflectometry refers to the measurement and analysis of optical signals transmitted in the optical fiber channel by using an optical time domain reflectometer. The backscattered Rayleigh light refers to the scattered light caused by the fluctuation of material density when the light is transmitted in the optical fiber channel. The phase delay change refers to the phase change in the process of optical signal transmission. The ultra-narrow linewidth pulsed laser refers to a device capable of generating pulsed laser with extremely narrow linewidth. The local light refers to a stable optical signal used as a reference. The photoelectric detector refers to a device for converting optical signals into electrical signals. The interference light intensity signal refers to the light intensity signal after the superposition of two lights. The improved LMS adaptive filtering algorithm refers to an optimized least mean square adaptive filtering method.

[0036] In the above embodiments, in the water supply pipe network system of a coastal city, due to the long-term influence of sea wind erosion and seawater backflow, some water supply pipes are seriously aged and leakage accidents occur frequently. In order to realize efficient and accurate leakage detection, the above detection scheme based on fiber vibration is adopted. In the key section of the urban water supply pipe network, a monitoring point is set every 200 meters, and a double-channel optical fiber sensor is deployed at each monitoring point. The first optical fiber channel is tightly coupled to the surface of the water supply pipe, so that it can accurately capture the pipe vibration signal; the second optical fiber channel maintains a preset distance of 10 centimeters from the water supply pipe, which is used to collect environmental noise signals. At a monitoring point, an ultra-narrow linewidth pulsed laser (wavelength 1550 nm, linewidth 5 kHz) generates an optical pulse, which is injected into the first optical fiber channel for optical time domain reflectometry. When the optical pulse is transmitted in the optical fiber, backscattered Rayleigh light is generated. At the same time, the local light generated by the laser and the backscattered Rayleigh light are superimposed in the photoelectric detector to obtain the interference light intensity signal. At this time, by analyzing the phase delay change of the backscattered Rayleigh light and the local light after interference, it can be preliminarily judged whether the pipe has abnormal vibration caused by fiber strain. For example, when a small leakage occurs in a section of pipe, the water flow impact will cause the pipe to vibrate, which in turn causes the optical fiber coupled to it to deform slightly, resulting in a change in the phase of the backscattered Rayleigh light.

[0037] In the above embodiment, the improved LMS adaptive filtering algorithm is used to process the interference light intensity signal and the environmental noise signal collected by the second channel. The traditional LMS algorithm is prone to slow convergence speed and poor filtering effect under complex and variable environmental noise. The improved algorithm introduces a variable step size mechanism to dynamically adjust the step size according to the real-time changes of the signal. When the noise intensity suddenly increases, the algorithm quickly increases the step size to speed up the suppression of noise; when the signal tends to be stable, the step size is reduced to improve the filtering accuracy, and finally the pure leakage vibration signal is obtained. From the obtained leakage vibration signal, time domain and frequency domain features are extracted respectively. In the time domain, peak value, rise time, pulse width and other leakage impact characteristics are extracted, which can intuitively reflect the intensity and duration of the leakage at the moment of occurrence. In the frequency domain, the leakage voiceprint feature is obtained through fast Fourier transform (FFT), and different types and degrees of leakage will produce unique frequency domain distribution, such as small crack leakage may have energy concentration in high frequency band, while large hole leakage has obvious energy characteristics in low frequency band. At the same time, according to the phase delay change obtained before, combined with the physical parameters of the optical fiber, the fiber strain feature of the first optical fiber channel is determined. For example, through a specific calculation formula, the phase change is converted into the strain value of the optical fiber, which reflects the influence degree of pipeline vibration on the optical fiber. The fiber strain feature, leakage impact feature and leakage voiceprint feature are fused in multiple dimensions. The principal component analysis (PCA) combined with artificial neural network can be used, first the multi-dimensional features are processed by PCA to remove redundant information and retain key features, and then artificial neural network is used to further learn the complex relationship between features, and finally the pipeline leakage feature vector containing key information of pipeline leakage is obtained.

[0038] In the above embodiments, in addition to the optical fiber vibration signal monitoring, an optical fiber temperature sensor and an optical fiber pressure sensor can be additionally deployed at the monitoring point. When a pipeline leaks, the water temperature and water pressure near the leakage point will change, and these changes are collected by the optical fiber temperature and pressure sensors and fused with the features extracted from the optical fiber vibration signal. For example, when the vibration signal shows that there may be a leak, and at the same time the temperature sensor detects a local drop in water temperature and the pressure sensor detects abnormal fluctuations in pressure, the three confirm each other, which can more accurately determine the occurrence of a leak and reduce the false alarm rate. Due to the difficulty of obtaining actual pipeline leakage data and the limited number of samples, a generative adversarial network (GAN) is used to generate virtual pipeline leakage data. By training the GAN, it learns the feature distribution of the real leakage data and generates a large number of simulated leakage data. These simulated data are combined with the actually collected data to expand the training data set, which is used to optimize the subsequent leakage detection model and improve the model's generalization ability and adaptability to different leakage scenarios. Considering that the optical fiber sensor may drift in performance due to environmental factors (such as temperature and humidity changes) during long-term use, a self-calibration dynamic compensation mechanism is designed. The sensor is calibrated regularly, and the differences between the standard signal and the sensor-acquired signal are compared to automatically adjust the sensor's parameters. At the same time, based on environmental monitoring data (such as real-time temperature and humidity), the collected signals are dynamically compensated to ensure that the sensor works stably and accurately in the long term.

[0039] In an optional embodiment, the pipeline leakage feature vector is input into the CNN-LSTM hybrid model to obtain pipeline leakage information output by the CNN-LSTM hybrid model after the CNN-LSTM hybrid model performs feature analysis and prediction processing on the pipeline leakage feature vector, specifically including: inputting the pipeline leakage feature vector into the CNN-LSTM hybrid model to enable the CNN-LSTM hybrid model to perform the following operations: the CNN-LSTM hybrid model, upon determining that the pipeline leakage feature vector is received, performs multi-scale time series modeling processing on the pipeline leakage feature vector to obtain an optimized feature vector; and the CNN-LSTM hybrid model performs multi-task joint prediction processing on the optimized feature vector to obtain the pipeline leakage information.

[0040] In the above embodiments, the CNN-LSTM hybrid model refers to a deep learning model combining a convolutional neural network (CNN) and a long short-term memory network (LSTM), the CNN is used to extract spatial features, and the LSTM is used to process time series data; the multi-scale time series modeling processing means feature extraction and time series analysis of the feature vector at different time scales; the optimized feature vector refers to a feature representation that fuses spatial and temporal information after multi-scale processing; the multi-task joint prediction processing means that the model simultaneously completes the prediction of multiple related tasks, such as leakage probability and leakage level prediction; and the pipeline leakage information is used to represent the comprehensive detection results including leakage probability, level, etc.

[0041] In the above embodiment, the fused pipeline leakage feature vector is input into the CNN-LSTM hybrid model. The CNN-LSTM hybrid model performs multi-scale time series modeling on the feature vector: three different size convolution kernels are used to process the feature vector in parallel, a small convolution kernel (3x1) captures short-term burst features such as high-frequency vibrations at the moment of leakage, a medium convolution kernel (11x1) extracts medium-term pattern features such as periodic vibrations caused by water flow fluctuations, and a large convolution kernel (19x1) obtains long-term trend features such as chronic leakage caused by pipeline aging. A double-layer bidirectional LSTM network is used to model the time series of multi-scale features, the forward LSTM learns the time series dependence from the past to the present, and the backward LSTM captures the time series information from the present to the future, and through the attention mechanism, the weights of different time steps are automatically allocated, for example, when processing the feature vector of a certain monitoring point, the model finds through multi-scale analysis that there are high-frequency vibration features in the short term, periodic fluctuations in the medium term, and a gradually increasing trend in the long term, and the comprehensive judgment is that there may be a developing leakage fault.

[0042] In the above embodiment, the CNN-LSTM hybrid model performs task decomposition on the optimized feature vector: leakage detection task, to determine whether there is a leak (binary classification); leakage positioning task, to determine the leakage location (regression prediction); and leakage degree evaluation, to evaluate the severity of the leak (multi-classification). The optimized feature vector is input into a shared fully connected layer, a Swish activation function is used to enhance the non-linear expression ability, and a Dropout layer is added to reduce overfitting. The leakage detection layer uses a Sigmoid activation function to output the leakage probability, the leakage positioning layer uses a linear activation function to predict the leakage position coordinates, and the leakage degree evaluation layer uses a Softmax activation function to output the severity probability distribution. According to the pipeline operation state, dynamically adjust the weights of each task, increase the weight of the leakage detection task during the water peak period, and increase the weights of the positioning and degree evaluation during the night low period, for example, in a certain prediction, the model outputs: leakage probability 0.92 (high risk), leakage position 235 meters away from the monitoring point, and severity "medium", suggesting immediate investigation.

[0043] In the above embodiment, the topology of the water supply network is constructed as a graph neural network, and the feature vectors of each monitoring point are taken as node features, and the pipeline connection relationship is taken as an edge. Through the graph convolution network (GCN), the spatial dependency between monitoring points is learned, and then combined with the LSTM to process the time sequence information to form a spatio-temporal graph neural network. This structure can capture the mutual influence between distant monitoring points in the pipe network, and improve the detection capability of complex leakage scenarios. The same CNN-LSTM model is deployed in the water supply network systems of multiple cities, and collaborative training is carried out using the federated learning framework. Each city uses local data to train the model, and only uploads model parameter updates rather than raw data, protecting privacy while sharing knowledge. Through federated learning, the model can learn the common features of different city pipe networks while retaining adaptability to local characteristics, significantly improving cross-regional leakage detection performance. The Monte Carlo Dropout technique is introduced to quantify the uncertainty of model prediction, calculating the confidence of each prediction result. When the prediction uncertainty exceeds the threshold, the active learning mechanism is triggered, automatically labeling these samples and requesting manual verification, and adding the verified data to the training set. This way can efficiently utilize limited labeling resources and gradually improve the model's performance on difficult samples. On the basis of multi-task prediction, a causal reasoning model is introduced to analyze the leakage causes. A causal graph of the water supply network is constructed, including pipeline material, service life, pressure change, environmental factors and other nodes. When a leak is detected, the most likely cause is traced back through causal reasoning, such as "pipeline aging" or "pressure anomaly", providing a more comprehensive basis for maintenance decisions.

[0044] In an optional embodiment, the CNN-LSTM hybrid model determines that a pipeline leakage feature vector is received, and performs multi-scale time sequence modeling processing on the pipeline leakage feature vector to obtain an optimized feature vector, specifically including: the CNN-LSTM hybrid model uses three groups of parallel convolution kernels of different scales to perform multi-scale feature extraction on the pipeline leakage feature vector to obtain local time-frequency features; the CNN-LSTM hybrid model performs max-pooling operation on the local time-frequency features to obtain dimension-reduced features; the CNN-LSTM hybrid model uses a two-layer bidirectional LSTM network to perform time sequence modeling on the dimension-reduced features to obtain time sequence correlation features, wherein the two-layer bidirectional LSTM network includes a first LSTM layer for forward propagation and a second LSTM layer for backward propagation; the CNN-LSTM hybrid model uses a self-attention layer to perform weighted processing on the time sequence correlation features to obtain weight features; and the CNN-LSTM hybrid model uses a fully connected layer to perform nonlinear mapping on the weight features to obtain the optimized feature vector.

[0045] In the above embodiments, parallel convolution kernels represent multiple sets of different scale convolution kernels (such as 1x3, 1x11, 1x19, etc.) working simultaneously in the CNN to extract multi-scale features; local time-frequency features refer to the frequency distribution characteristics of the signal in the local time window; the maximum pooling operation refers to taking the maximum value of the feature map to reduce the dimension and retain the key features; the reduced dimension feature refers to the low-dimensional feature representation after pooling; the bidirectional LSTM network refers to a network composed of forward and backward LSTM layers, which can capture past and future time series information simultaneously; the time series correlation feature represents the dependence relationship characteristics of the signal in the time series; the self-attention layer refers to a mechanism for weighting features by calculating the correlation between features; the weight feature refers to the feature after attention weighting; and the fully connected layer refers to a network layer in which all neurons are connected, used for nonlinear mapping.

[0046] In the above embodiments, in the monitoring of the water supply network in the above coastal city, the pipeline leakage feature vector obtained by fusing multi-dimensional features is input into the CNN-LSTM hybrid model. The CNN-LSTM hybrid model starts three sets of parallel convolution kernels of different scales, and the small convolution kernel is sensitive to capture short-term local features such as high-frequency vibrations generated at the moment of sudden pipeline leakage; the medium convolution kernel focuses on analyzing periodic vibration and other medium-term pattern features caused by unstable water flow; and the large convolution kernel focuses on mining long-term trend features of chronic leakage caused by pipeline aging, thereby obtaining rich local time-frequency features. The maximum pooling operation processes the local time-frequency features, reduces the data dimension while retaining the key information, and generates reduced dimension features. The reduced dimension features are sent to two layers of bidirectional LSTM network, the forward LSTM layer learns the dependence relationship between features from the past to the present time dimension, and the reverse LSTM layer captures the time series information from the present to the future, both of which cooperate to model the features in depth time series, and obtain features containing rich time series correlation. The self-attention layer weights the time series correlation features, assigns weights according to the importance of the features at different time steps for judging the leakage, highlights the key information, and forms the weight features. The fully connected layer implements nonlinear mapping on the weight features, and through the complex calculation of multiple neurons, it is converted into the final optimized feature vector, providing accurate basis for subsequent judgment of pipeline leakage.

[0047] In the above embodiment, an adaptive convolution adjustment strategy is set during the running of the CNN-LSTM hybrid model. The fluctuation, frequency distribution and complexity of the input feature vector are analyzed in real time. If it is detected that the high-frequency components in the feature vector increase, it is suspected that there is a sudden leakage, and the weight of the large convolution kernel in multi-scale feature extraction is automatically increased to strengthen the capture of high-frequency details. If it is found that the feature changes are relatively flat, there may be chronic leakage, and the weight proportion of the large convolution kernel is increased. At the same time, a hierarchical attention strengthening mechanism is introduced to build a special connection between the two layers of bidirectional LSTM network. The feature information output by the first layer of LSTM is calculated and filtered to form a context vector. This vector and the output of the first layer are used as the input of the second layer of LSTM, so that the second layer network can focus more on the features that are important to the judgment of leakage, and the sensitivity of the model to key leakage information is improved. A dynamic time adaptation scheme is designed. The CNN-LSTM hybrid model can automatically identify the periodicity of the signal according to the autocorrelation characteristics of the input feature, and then adjust the time step of the LSTM network to process the data. When detecting a leakage vibration signal with obvious periodicity, the time step is adjusted to a value matching the signal period, so that the CNN-LSTM hybrid model can more efficiently capture the change pattern of the leakage signal. An anti-disturbance training link is also added. During the training of the CNN-LSTM hybrid model, different degrees of simulated noise and interference are added to the input data, so that the CNN-LSTM hybrid model can learn to accurately extract leakage features in a complex interference environment, and enhance the anti-interference ability and detection stability of the model in actual complex working conditions. Even in the face of strong environmental noise, the pipeline leakage situation can be accurately judged.

[0048] In an optional embodiment, the CNN-LSTM hybrid model performs multi-task joint prediction processing on the optimized feature vector to obtain pipeline leakage information, specifically including: the CNN-LSTM hybrid model inputs the flattened optimized feature vector into a shared fully connected layer for first nonlinear mapping to obtain shared features; the CNN-LSTM hybrid model inputs the shared features into a first task-specific layer for second nonlinear transformation to obtain first task features; the CNN-LSTM hybrid model performs binary classification processing on the first task features using a Sigmoid activation function to output a pipeline leakage probability, wherein the pipeline leakage information includes the pipeline leakage probability; the CNN-LSTM hybrid model inputs the shared features into a second task-specific layer for third nonlinear transformation to obtain second task features; and the CNN-LSTM hybrid model performs multi-classification processing on the second task features using a Softmax activation function to output a pipeline leakage grade, wherein the pipeline leakage information includes the pipeline leakage grade.

[0049] In the above embodiments, the flattening operation refers to converting a multi-dimensional feature vector into a one-dimensional vector for processing by a fully connected layer; the shared fully connected layer refers to a fully connected layer used by multiple tasks in multi-task learning to extract general features; the shared feature refers to general feature representation; the task-specific layer refers to a network layer designed specifically for different tasks (such as binary classification, multi-classification); the nonlinear transformation refers to introducing a nonlinear relationship through an activation function (such as a ReLU function, etc.); the Sigmoid activation function refers to a function that maps the output to the [0, 1] interval, suitable for binary classification; the Softmax activation function refers to a function that maps the output as a probability distribution, suitable for multi-classification; the pipeline leakage probability is used to represent the numerical value of the possibility of pipeline leakage; the pipeline leakage level refers to dividing the severity of leakage into multiple levels (such as slight, moderate, and severe).

[0050] In the above embodiments, in the water supply network monitoring system of the above coastal city, after the CNN-LSTM hybrid model completes the multi-scale time series modeling of the pipeline leakage feature vector and obtains the optimized feature vector, it enters the prediction decision stage. The optimized feature vector is flattened and then input into the shared fully connected layer. This layer acts like an "information filter" that filters out common key information related to pipeline leakage through the first nonlinear mapping from complex features, forming shared features that contain basic clues for judging whether the pipeline is leaking, locating the leakage position, and evaluating the leakage level. The shared features are divided into two paths. One path enters the first task-specific layer, where the second nonlinear transformation is performed, further mining and strengthening the relevant features for the task of judging whether the pipeline is leaking, to obtain the first task feature. The first task feature is processed using the Sigmoid activation function, which maps the feature value to between 0 and 1, outputting a specific numerical value as the pipeline leakage probability. For example, when the value is close to 1, it indicates a high probability of pipeline leakage; if the value is close to 0, it means that the pipeline has little risk of leakage. The other path of shared features enters the second task-specific layer, which undergoes a third nonlinear transformation to obtain the second task feature specifically for evaluating the pipeline leakage level. Then, through the Softmax activation function, the second task feature is converted into a probability distribution of each leakage level, thereby outputting the pipeline leakage level. For example, the output result may show a slight leakage probability of 0.2, a moderate leakage probability of 0.6, and a severe leakage probability of 0.2, from which the severity of pipeline leakage can be clearly determined.

[0051] In the above embodiments, a dynamic threshold adjustment mechanism is introduced in the shared fully connected layer. The threshold output by the shared fully connected layer can be dynamically adjusted according to the real-time operation state of the water supply network, such as the water use peak period, the low valley period at different times, and the distribution rule of historical leakage data. During the water use peak period, due to the complex change of pipeline pressure, the threshold for judging the leakage probability can be increased to avoid misjudgment caused by pressure fluctuation; and during the water use low valley period, the threshold is appropriately reduced, and any subtle leakage signs are not missed. The feature interaction module is set between the task-specific layers, so that the features generated by the first task-specific layer and the second task-specific layer can interact with each other. For example, some feature information obtained when judging whether to leak will be transmitted to the second task-specific layer for evaluating the leakage level to assist it to more accurately judge the leakage level; vice versa. This feature interaction breaks the information barrier between tasks, enabling the model to share key clues when predicting different tasks, improving the accuracy and relevance of prediction. At the same time, the parameters of the activation function are adaptively adjusted. The Sigmoid function and the Softmax function are not fixed, but automatically adjust their own parameters according to the distribution of input features. When the fluctuation of input features is large, the Sigmoid function will adjust the slope of the curve, making it more sensitive to small changes, so as to more accurately output the leakage probability; the Softmax function will optimize the calculation method of probability distribution according to the difference degree of different leakage level features, to ensure that the judgment of leakage level is more reasonable and reliable.

[0052] In an optional embodiment, after the CNN-LSTM hybrid model utilizes a Softmax activation function to perform multi-classification processing on the second task features to output the pipeline leakage level, the method further comprises: extracting a standard leakage voiceprint feature corresponding to the pipeline leakage level from a pre-established leakage voiceprint feature library; matching the leakage voiceprint feature in the pipeline leakage feature vector with the standard leakage voiceprint feature to obtain a pipeline leakage type; in a case where it is determined that the pipeline leakage type is a micro-leakage type, determining a low-frequency band energy proportion in the leakage voiceprint feature to obtain a low-frequency band energy proportion result; in a case where it is determined according to the low-frequency band energy proportion result that the low-frequency band energy proportion is greater than a first preset threshold, determining that the pipeline leakage cause is aging of an interface rubber sealing element; in a case where it is determined according to the low-frequency band energy proportion result that the low-frequency band energy proportion is less than or equal to the first preset threshold, determining that the pipeline leakage cause is a pipeline weld micro-pore; in a case where it is determined that the pipeline leakage type is a medium-leakage type, determining a medium-frequency band energy distribution in the leakage voiceprint feature to obtain a medium-frequency band energy distribution result; in a case where it is determined according to the medium-frequency band energy distribution result that a standard deviation of the medium-frequency band energy distribution is greater than a second preset threshold, determining that the leakage cause is a pipeline corrosion pit; in a case where it is determined that the pipeline leakage type is a serious-leakage type, detecting whether there is a high-frequency burst peak value in the leakage voiceprint feature; in a case where it is detected that there is a high-frequency burst peak value in the leakage voiceprint feature, and an amplitude of the high-frequency burst peak value is greater than a third preset threshold, determining that the leakage cause is a pipe body rupture; performing fault propagation prediction according to the pipeline leakage cause to obtain a leakage expansion risk level; and generating a priority-ordered repair strategy set according to the pipeline leakage level and the leakage expansion risk level.

[0053] In the above embodiments, the pre-established leakage voiceprint feature library refers to a pre-established voiceprint feature database containing different leakage types and levels; the standard leakage voiceprint feature refers to a typical voiceprint feature corresponding to a leakage level; the pipeline leakage type refers to a category classified according to leakage features (such as micro-leakage, medium-leakage, and serious-leakage); the low-frequency band energy proportion is used to represent a proportion of energy of a low-frequency part (such as 0-50 Hz, etc.) in the leakage voiceprint in total energy; the medium-frequency band energy distribution refers to an energy distribution feature of a medium-frequency part (such as 50-500 Hz, etc.); the high-frequency burst peak value refers to an instantaneous energy spike of a high-frequency band (such as 500 Hz or above, etc.); the fault propagation prediction refers to predicting a leakage development trend according to a leakage cause; the leakage expansion risk level refers to an assessment of a harm degree possibly caused by the leakage; and the repair strategy set refers to a priority list of maintenance schemes generated according to the leakage level and the risk.

[0054] In the above embodiment, in the daily monitoring of the water supply pipe network in a certain coastal city, the pipe leakage feature vector obtained based on fiber vibration collection and multi-dimensional feature fusion is input into the subsequent analysis process. The standard leakage voiceprint feature corresponding to the current detected pipe leakage level is quickly extracted from the preset leakage voiceprint feature library. The standard leakage voiceprint feature and the leakage voiceprint feature in the pipe leakage feature vector are matched in detail, and when it is determined that the pipe leakage type is micro leakage, the low frequency band of the leakage voiceprint feature is focused on, and the energy proportion is accurately calculated. When the calculated low frequency band energy proportion exceeds the first preset threshold, it is judged that the cause of the pipe leakage is the aging of the interface rubber sealing element. Because in the long-term sea wind erosion and seawater backflow environment, the rubber sealing element is easy to accelerate aging, so that a small gap appears at the interface, causing the vibration signal in the low frequency band to present obvious energy concentration characteristics. If the low frequency band energy proportion is less than or equal to the first preset threshold, it is determined that the leakage cause is the pipe weld micro-pore, which is due to the existence of the micro-pore at the weld, the vibration signal generated under the action of water flow pressure has relatively weak low frequency band energy.

[0055] In the above embodiment, assuming that the detection determines that the pipe leakage type is medium leakage, the energy distribution of the medium frequency band of the leakage voiceprint feature is analyzed. When it is found that the standard deviation of the medium frequency band energy distribution is greater than the second preset threshold, it means that the energy distribution fluctuation is large, and in combination with the actual working condition, it is judged that the leakage is caused by pipe corrosion pits. With the passage of time, the pit formed by the corrosion of the pipe continuously expands, and the vibration generated by the water flow impacting the inner wall of the pit presents unstable energy distribution in the medium frequency band. If the pipe leakage type is detected as serious leakage, it will be carefully detected whether there is a high frequency burst peak value in the leakage voiceprint feature. Once the high frequency burst peak value is captured and its amplitude exceeds the third preset threshold, it is immediately determined that the leakage cause is pipe body rupture. Under the impact of strong water flow, the pipe body rupture will produce strong vibration in an instant, and this vibration appears in the form of high frequency burst peak value in the voiceprint feature.

[0056] In the above embodiment, after determining the cause of the pipe leakage, the fault propagation is predicted according to the actual situation of the pipe material, service life, surrounding water pressure and the like. Through comprehensive evaluation, the leakage expansion risk level is obtained. For example, for the serious leakage caused by pipe body rupture, considering that the surrounding water pressure is large and the pipe material has aged, the leakage expansion risk level is evaluated as high. Finally, according to the pipe leakage level and the leakage expansion risk level, a set of repair strategies sorted by priority is generated. For the serious leakage with high risk, a professional repair team is arranged to carry emergency pipe materials to the scene for emergency plugging and replacement; for the micro leakage with low risk, a detailed maintenance plan is made to arrange sealing element replacement or weld repair work during the non-water peak period.

[0057] In the above embodiments, a dynamic weight adjustment strategy is introduced in the matching process. According to different time periods (such as daytime water peak, nighttime water valley), different environmental conditions (such as rainstorm weather, high temperature weather), the weight of each dimension feature in the matching of the pipeline leakage feature vector and the standard leakage voiceprint feature is automatically adjusted. For example, in rainstorm weather, environmental noise may interfere with the voiceprint feature, at this time, the weight of the feature affected by noise is reduced, and the weight of the relatively stable feature is increased, so as to more accurately determine the leakage type. In the fault propagation prediction link, the geographical information around the pipeline is combined. If there are subway construction, road excavation and other engineering activities near the pipeline, these external factors will be taken into account to re-evaluate the leakage expansion risk level. For example, if the pipeline body is broken near the pipeline, the vibration generated by the construction may accelerate the damage of the pipeline, and the system will correspondingly increase the leakage expansion risk level, and prioritize the protection measures in the repair strategy set to avoid the mutual influence of construction and leakage to cause more serious accidents. When generating the repair strategy set, the real-time scheduling of the maintenance resources is considered. When the number of maintenance teams is limited, the priority order of the repair strategy is optimized according to the distance between each leakage point and the maintenance base and the inventory situation of the required maintenance materials. The leakage points with short distance and sufficient maintenance materials are preferentially selected for repair, and the maintenance resources in other areas are coordinated to support the leakage points in emergency and resource shortage, thereby improving the overall maintenance efficiency.

[0058] In an optional embodiment, the double-channel optical fiber vibration signals are synchronously sampled by using an FPGA master clock to obtain time-aligned pipeline vibration signals and environmental noise signals, and the method specifically comprises the following steps: controlling the FPGA master clock to send a Sync message to a vibration collection terminal and recording a first sending timestamp of the FPGA master clock sending the Sync message; recording a first receiving timestamp of the vibration collection terminal receiving the Sync message under the condition that the vibration collection terminal receives the Sync message; controlling the vibration collection terminal to return a Delay_Req message to the FPGA master clock and recording a second sending timestamp of the vibration collection terminal returning the Delay_Req message; recording a second receiving timestamp of the FPGA master clock receiving the Delay_Req message under the condition that the FPGA master clock receives the Delay_Req message; calculating a link delay time according to the first sending timestamp, the first receiving timestamp, the second sending timestamp and the second receiving timestamp, wherein the link delay time is used to represent the signal transmission delay between the FPGA master clock and the vibration collection terminal; performing time compensation processing on the initial pipeline vibration signals and the initial environmental noise signals according to the link delay time, and performing interpolation alignment processing on the compensated initial pipeline vibration signals and the initial environmental noise signals to obtain the time-aligned pipeline vibration signals and the environmental noise signals.

[0059] In the above embodiment, the Sync packet refers to a synchronization signal packet sent by the FPGA master clock, used to trigger sampling and record the timestamp; the Delay_Req packet refers to a delay request packet returned by the vibration collection terminal, used to measure the round-trip time of the signal; the first sending timestamp indicates the time when the FPGA master clock sends the Sync packet; the first receiving timestamp indicates the time when the terminal receives the Sync packet; the second sending timestamp indicates the time when the terminal returns the Delay_Req packet; the second receiving timestamp indicates the time when the FPGA master clock receives the Delay_Req packet; the link delay time refers to the transmission delay of the signal between the FPGA master clock and the vibration collection terminal; the time compensation processing refers to the time offset correction of the sampling data according to the delay time; the interpolation alignment processing refers to the time axis alignment of the compensated signal through the interpolation algorithm, ensuring that the sampling points strictly correspond. In the above coastal city water supply network monitoring system, the accurate time synchronization between the vibration collection terminal distributed in each monitoring point and the FPGA master clock is the key to ensure effective data collection. The FPGA master clock sends the Sync packet to the vibration collection terminal first, and at the same time, the high-precision timer of the FPGA master clock immediately records the sending time, generating the first sending timestamp. As soon as the vibration collection terminal receives the Sync packet, it responds quickly, and the internal time recording module immediately records the receiving time, forming the first receiving timestamp. The vibration collection terminal returns the Delay_Req packet to the FPGA master clock, and records the second sending timestamp of the return. After receiving the Delay_Req packet, the FPGA master clock records the second receiving timestamp again through the timer. This series of operations completely records the key time nodes of the signal transmission between the two. Subsequently, according to the four timestamps, the link delay time is accurately calculated through specific calculation logic, which accurately reflects the delay condition of the signal transmission between the FPGA master clock and the vibration collection terminal. With the link delay time, the initial pipeline vibration signal and the initial environmental noise signal can be time compensated. The time deviation caused by transmission delay is corrected, and then through interpolation alignment processing, the pipeline vibration signal and the environmental noise signal with completely aligned time are finally obtained, laying a reliable foundation for subsequent signal analysis.

[0060] In the above embodiment, a dynamic link monitoring mechanism is introduced. The trend of link delay time is continuously monitored. Once abnormal fluctuation of delay time is found, such as a large increase in a short time, more frequent interaction of Sync message and Delay_Req message is automatically triggered to obtain more accurate real-time link delay time. At the same time, the parameters of time compensation and interpolation alignment are dynamically adjusted in combination with the environmental factors around the water supply network, such as whether there is large-scale construction near the electromagnetic interference, to ensure that the accurate time alignment of signals can be realized in complex environment. A multi-link redundancy backup strategy can also be added. Multiple communication links are set at each monitoring point to connect the vibration collection terminal and the FPGA master clock. When the link delay time of the main link exceeds a certain threshold, the system automatically switches to the backup link with lower and more stable delay, ensuring the timeliness and stability of signal transmission. After switching the link, the system quickly calculates the delay time of the new link and adjusts the signal time compensation to ensure that data collection is not affected by link switching.

[0061] It should be noted that the above described embodiments are only part of the embodiments of the present application, not all. The present application will be specifically described below in conjunction with specific embodiments.

[0062] The present application provides a system architecture, Figure 2 is a layered architecture diagram of the optical fiber vibration auscultation system in the embodiments of the present application, please refer to Figure 2 The system is composed of three parts:

[0063] Vibration sensing layer (obtain the converted electrical signal of Rayleigh scattering light through the Internet of Things optical fiber, the laser wavelength and the refractive index of the optical fiber are fixed values): Install two optical fibers on the water supply network path. One is a vibration optical fiber attached to the surface of the pipeline, which is laid along the pipeline path and packaged with high-pressure resistant silicone (pressure resistance ≥2MPa) to tightly adhere to the surface of the pipeline and sense the 0.1-200Hz low-frequency vibration excited by the leakage water pressure. The other is laid overhead 10cm (of course, it can also be 9cm, 10.5cm, 11cm, etc., which is not limited here) away from the pipeline to collect environmental noise and offset common-mode interference through an improved LMS adaptive filtering algorithm. Based on FPGA, μs-level time synchronization is realized to ensure the time domain alignment of vibration-noise signals.

[0064] FPGA as the master clock sends Sync message to the vibration / noise collection terminal through the optical fiber network, records the sending time stamp The device (such as vibration sensor) records the receiving time stamp And replies Delay_Req message, records the reply time stamp FPGA records the receiving time stamp The link delay is calculated by the following formula :

[0065]

[0066] Compensate time deviation, interpolate and align vibration and noise signals by the following formula:

[0067]

[0068] Wherein, is the aligned signal sequence, representing the signal value after time compensation and interpolation processing; is the original sampling signal sequence, representing the vibration or noise signal value without processing; is the index of the sampling point, used to traverse the original signal sequence; is the new sampling point index, representing the time point of the aligned signal; is the time deviation, representing the amount of time delay that needs to be compensated; is the sampling interval, representing the time interval between adjacent sampling points. Sinc is the sinc function, used for signal interpolation reconstruction.

[0069] Data extraction layer (the collected signals are data cleaning, error data is identified and removed; pre-processing provides standardized data for subsequent analysis; considering flexibility and analysis ability, data is stored; data management ensures data security, accessibility and efficient use):

[0070] Extract multi-dimensional features of vibration signals, analyze leakage impact strength and frequency through time domain feature analysis, analyze leakage voiceprint spectrum characteristics through frequency domain feature analysis, and analyze phase changes caused by optical fiber strain through optical characteristics.

[0071] Analysis decision layer (feature recognition is performed through algorithm, bidirectional LSTM is stacked to learn the correlation features before and after the leakage event, nonlinear mapping is performed through full connection layer, sigmoid activation function is used, and binary classification probability is output):

[0072] AI feature fusion, construct CNN-LSTM hybrid neural network, input layer is multi-source feature vector, output layer is leakage probability and leakage level, and training data set contains typical working condition samples.

[0073] Figure 3 It is a hardware composition schematic diagram of the optical fiber vibration auscultation system in the embodiment of the application, refer to Figure 3The hardware components include: a laser emitting end for generating high-precision laser pulse signals, controlling the wavelength, power and pulse characteristics of the laser; a distributed optical fiber for collecting vibration signals by sensing changes in the vibration, sound and other changes of the water supply pipeline to monitor the state of the water supply pipeline; a water supply pipeline as the monitoring object for conveying and distributing water, where possible leaks, damage and other problems will generate characteristic vibration and acoustic signals, etc. The characteristics of the pipeline such as material, pressure, flow rate, etc. will affect the characteristics of the vibration signal; a spectrum analyzer for receiving scattered light signals returned from the optical fiber, analyzing the spectral characteristics of the light signal, processing Rayleigh scattering signals and extracting vibration information; an optical-electricity converter for converting optical signals into electrical signals, performing signal conditioning and preprocessing to provide standard electrical signal input for subsequent analysis; a stethoscopic system (which is a terminal analysis platform in the optical fiber vibration stethoscopic system) for receiving signals converted by the optical-electricity converter, performing signal processing and analysis, realizing leak detection and positioning, providing data display and early warning functions. Signal flow: distributed optical fibers are laid along the water supply pipeline network, a periodic pulse is emitted to the optical fiber using a pulse laser emitter, scattered light (Rayleigh) information is obtained through an optical wave analyzer, optical signals are converted into electrical signals through an optical-electricity converter, and the electrical signals are provided to the stethoscopic system for leak diagnosis business application.

[0074] Figure 4 is a functional module schematic diagram of the optical fiber vibration stethoscopic system in the embodiment of the present application, refer to Figure 4 The functional module includes:

[0075] A data acquisition module acquires signal data on site and transmits the data to the stethoscopic system in the form of electrical signals.

[0076] A data extraction module analyzes and filters signal characteristics and identifies key features.

[0077] An analysis and identification module identifies normal and different leak states.

[0078] An alarm reaching module monitors and analyzes the identification results, generates alarm information if there is an anomaly, and pushes the alarm information to the alert through a reaching platform; Figure 5 is a functional schematic diagram of the alarm reaching module in the embodiment of the present application, refer to Figure 5The module includes three sub-functions: alarm monitoring trigger function: real-time monitoring of auscultation system data, setting alarm threshold and trigger condition, automatically triggering alarm mechanism when detecting abnormal data, ensuring that abnormal conditions can be discovered in time. Alarm grading analysis function: grading the severity of alarm events (such as critical, important, and ordinary levels), analyzing leakage size and potential impact, evaluating processing priority, and providing processing suggestions of different levels. Alarm notification function: according to the alarm level, different notification methods (such as SMS, email, APP push, etc.) are selected, and the alarm information is sent to the relevant person in charge. Ensure that information is sent to the corresponding processing personnel in time, and track the alarm processing status.

[0079] Diagnosis analysis module: analyze the causes of leakage according to the leakage situation; Figure 6 is a functional diagram of the diagnosis analysis module in the embodiments of the present application, see Figure 6 The module includes three sub-functions: running anomaly monitoring function: monitoring the running state of the water supply pipeline, identifying abnormal conditions deviating from normal running parameters, tracking changes in vibration, acoustic and other signals of the water supply pipeline in real time, and determining the time and location of the anomaly. Diagnosis result analysis function: in-depth analysis of the detected water supply pipeline anomaly, judging the type and severity of the leakage, evaluating the impact of the leakage on water supply, and generating a diagnosis report. Event root cause analysis function: trace the causes of water supply pipeline leakage, analyze the influence of pipeline material, water pressure, flow and other factors, and determine the specific cause of the leakage, providing basis for subsequent maintenance and prevention.

[0080] Suggestion decision module: recommend corresponding corrective measures according to the leakage cause.

[0081] Data visualization module: visual display of abnormal operation data and system diagnosis result display.

[0082] The embodiments of the present application also provide a fiber vibration auscultation process for water supply pipeline network leakage, which includes the following steps:

[0083] Step one, based on phase-sensitive optical time domain reflection (Φ-OTDR) to realize water supply pipeline network leakage detection, Figure 7 is a principle diagram of fiber Rayleigh scattering measurement in the embodiments of the present application, see Figure 7, demonstrates the core principle of phase-sensitive optical time-domain reflectometry (Φ-OTDR) in water supply pipeline leak detection, i.e. a narrow-linewidth pulsed laser is emitted to the sensing fiber by a pulsed laser, and the light propagates in the fiber and collides with the molecular structure to produce Rayleigh scattering; when the pipeline leaks, the water pressure vibration at the leakage point is transmitted to the fiber close to the pipeline through the pipe wall, causing local strain in the fiber; according to the photoelastic effect, the strain causes the refractive index of the fiber to change, and then modulates the phase of the backscattered Rayleigh light; the backscattered light carrying the phase change information returns along the original path, and the phase difference Δφ is converted into an interference light intensity fluctuation signal through differential interference between the photoelectric detector and the local light; the analyzer accurately calculates the leakage position according to the time difference between the light pulse emission and the scattered light return, and analyzes the vibration characteristics in the interference light intensity signal, realizing high-sensitivity identification and positioning of the leakage event:

[0084] When the water supply pipeline leaks, the vibration excited by the water pressure at the leakage point is transmitted to the sensing fiber through the pipe wall, causing local strain in the fiber. According to the photoelastic effect of the fiber, the strain changes the refractive index distribution of the fiber, and then causes the phase delay of the backscattered Rayleigh light to change. The amount of phase change and the strain satisfy the relationship:

[0085] wherein, is the wavelength of the laser, is the refractive index of the fiber, is the initial length of the fiber, is the change in the length of the fiber, is the change in the refractive index.

[0086] The differential delay interference technology is used to realize phase-intensity conversion, and an ultra-narrow linewidth pulsed laser is used, which has high coherence, so that the scattered light waves in the pulse area form stable interference. The backscattered Rayleigh light and the local light (with a matched delay time) are superimposed on the photoelectric detector, and the interference light intensity is:

[0087] wherein, is the interference light intensity, indicating the final measured light intensity signal, is the intensity of the scattered light field, is the intensity of the local oscillation light field, is the amplitude of the light field, is the cosine function of the phase difference.

[0088] The vibration position is determined according to the light pulse propagation time (T) , and the time-domain vibration signal is constructed by a continuous pulse sequence. The improved LMS adaptive filtering algorithm is used for noise suppression, and the ambient noise is collected by the suspended noise fiber to dynamically cancel the common-mode interference. The pipeline vibration feature library is pre-trained to support adaptive updating of cast iron / PE pipeline noise.

[0089] The output time series data of the Φ-OTDR is generated by measuring the phase and intensity changes of the back Rayleigh scattering light, and contains the following contents:

[0090] The back Rayleigh scattering light intensity sequence is a sequence of changes in the intensity of back scattering light generated by Rayleigh scattering when the probe light pulse propagates in the optical fiber, reflecting the scattering characteristics of each position of the optical fiber.

[0091] The phase change sequence is caused by the change of the refractive index or length of the optical fiber due to external vibration or strain, causing the phase fluctuation of the back Rayleigh scattering light The phase time sequence is obtained by coherent detection demodulation.

[0092] The time domain positioning information is the spatial position on the optical fiber corresponding to each data point in the time sequence, and the distance is calculated by the speed of light and the time delay. The spatial features of the target, such as the contour feature and the amplitude feature, can be extracted by using the CNN convolutional neural network. However, the traditional CNN has a fixed window function length, and the frequency resolution and time resolution cannot be well balanced. If the convolution kernel scale and span are too small, the time resolution of the signal is good, and it is sensitive to high-frequency feature changes, but it cannot learn the low-frequency features existing in the signal. On the contrary, a larger scale convolution kernel corresponds to a larger span, which can learn information in a longer time range, i.e. low-frequency features existing in the signal, but cannot well reflect the high-frequency characteristics. Therefore, the present application uses multiple one-dimensional convolution kernels of different scales to perform parallel convolution on the input information to extract signal features of different time scales.

[0093] In addition to spatial features, the output signal of the Φ-OTDR also has time sequence features, and the characteristic pulses on the curve are closely related to the time and the state before and after them. Therefore, on the basis of the CNN model, the LSTM model is further used to effectively extract the time sequence features of the signal. The LSTM controls the information flow through the gating mechanism, allowing the model to decide what information to retain, update, and forget. The cell state maintains long-term memory, and the sampling points within the time window are used as network inputs to obtain prediction values based on the sampling points within the window. The window is moved by a set step size, and the sequence is continuously predicted to extract the time sequence features of the signal.

[0094] Step two, Figure 8 is a structural diagram of the CNN-LSTM hybrid model in the embodiments of the present application, which is described in detail in Figure 8 The CNN-LSTM hybrid model combines the advantages of CNN and LSTM, and can obtain the contour feature, amplitude feature and time sequence feature of the signal at the same time. After feature fusion, the sensitivity and accuracy of the model are effectively improved, and through multi-task learning, the leakage / non-leakage event classification and leakage level prediction are realized:

[0095] 1) Input layer processing:

[0096] Top Input Layer ( The input is time-series data (i.e., the output signal of Φ-OTDR - time series). The time-series data undergoes initial feature extraction through two layers of CNN and Pooling. CNN Layer 1 (the first convolutional layer) extracts local spatial features from the input data using convolutional kernels (filters), identifying low-level features. Pooling Layer 1 downsamples the features output from the convolutional layers, reducing spatial dimensionality and thus computational cost. CNN Layer 2 (the second convolutional layer) further extracts higher-order spatial features, combining low-level features (such as edges) to form more complex patterns by stacking multiple convolutions. Pooling Layer 2 compresses features, further reducing parameters and improving the model's robustness to small changes in the input. Deep pooling captures more global spatiotemporal features. The fully-connected layer maps the pooling output features to the target space, integrating all features through a weight matrix to achieve non-linear transformation.

[0097] 2) LSTM timing modeling:

[0098] Two LSTM layers (LSTM Layer 1 and LSTM Layer 2); each LSTM layer processes sequence data, enabling forward and backward propagation; The hidden state of the LSTM is represented to record temporal information; the stacking of two LSTM layers enhances the model's ability to capture long-term dependencies.

[0099] 3) Self-attention layer:

[0100] In the circle of self-attention layer Represents the characteristics of each time step; This indicates the attention weight at each time step; the "+" sign indicates that the weighted features are merged.

[0101] 4) Multi-task output processing:

[0102] The output nodes of the bottom Output layer are actually divided into two groups:

[0103] A set of binary classifications for leaky / non-leaking functions (Sigmoid function activation);

[0104] Another set of multiple classifications for leakage levels (Softmax function activation).

[0105] The CNN feature extraction (extracting spatial features layer by layer (CNN→Pooling→CNN→Pooling)) is described in detail as follows:

[0106] Three groups of parallel convolution kernels (kernel one size 1×3, step 1; kernel two size 1×11, step 4; kernel three size 1×19, step 7) are used to extract local time-frequency features and capture the local correlation of high-frequency impact and low-frequency pressure fluctuation in the leakage signal. The large convolution kernel covers a wide time window of low-frequency pressure fluctuation and extracts long-term pressure fluctuation features. The medium convolution kernel focuses on medium-frequency transient events and analyzes the local mutation of the signal through medium-scale to enhance the time-frequency resolution. The small convolution kernel captures the high-frequency impact component and extracts microsecond-level transient features using a narrow window.

[0107] Cross-scale feature splicing: The outputs of the three groups of convolution are spliced along the channel dimension to form a multi-scale feature tensor , which is input into the LSTM layer to model the time sequence dependence. The SE (Squeeze-Excitation) module is used to dynamically allocate weights to each frequency band to enhance key features.

[0108] High-frequency impact detection: The envelope of the small kernel output is calculated by Hilbert transform, and if the envelope peak value exceeds 3 times the standard deviation of the baseline, a high-frequency event marker is triggered. The envelope mean (baseline_mean) is calculated as the baseline, and the envelope standard deviation (baseline_std) is calculated, and the trigger threshold baseline_mean+3baseline_std is set. The 3 times standard deviation threshold here is based on the "small probability event" principle in statistics, which considers that when the signal exceeds this threshold, it is an abnormal situation.

[0109] Low-frequency correlation analysis: The cross-correlation function of the large kernel output and the signal is calculated, and if the correlation coefficient is greater than 0.7 (of course, it can also be 0.8, 0.85, 0.9, etc., which is not limited here), it is determined as a leakage correlation event.

[0110] Max pooling (pooling window 2×1) is used to reduce the dimension and retain key features (peak value of high-frequency impact: amplitude of transient event, vertex of leakage pulse, extreme point of low-frequency waveform: key phase of pressure fluctuation, wave peak, trough, etc. These key features together constitute the identification benchmark of the leakage signal). Reduce the amount of calculation.

[0111] Figure 9 is an operation schematic diagram of the convolution kernel in the embodiment of the application, and Figure 9 shows the process of one-dimensional convolution operation:

[0112] 1) Input part:

[0113] Input signal: The input signal is a 2x3 matrix, with the upper row data being [a b c] and the lower row data being [xy z];

[0114] Filter parameters: kernel_size=1x3 (one-dimensional filter, size 3), stride=1 (step size 1, indicating that the filter moves 1 unit at a time).

[0115] 2) Convolution process:

[0116] The filter slides from left to right, calculating the product sum of the values in the current window and the filter each time. The dashed arrows indicate how the calculation results at each position are obtained.

[0117] 3) Calculation formula:

[0118] ; ; ; .

[0119] 4) Output result:

[0120] Result: A one-dimensional array [r1 r2 r3 r4] containing 4 elements is generated. It should be noted that the convolution kernel operation process is a prior art and will not be described here.

[0121] The following explains the LSTM time series modeling (used to receive the flattened feature sequence of the CNN output, modeling the dependency relationship):

[0122] Figure 10 is a structural diagram of the LSTM time series modeling in the embodiments of the present application, referring to Figure 10 , 2 layers of bidirectional LSTM (128 units) are stacked to capture long-term dependencies and solve the limitations of traditional LSTM unidirectional propagation, while learning the features of the pre- and post-leakage events. A self-attention layer (Self-Attention) is introduced to adaptively assign feature weights and enhance the significance of key time steps, improving the recognition ability of weak leakage signals. The bottom input layer: represents the time series data (i.e., the output signal-time series of Φ-OTDR); the middle L1 layer is the first LSTM layer; the upper L2 layer is the second LSTM layer; and the top output layer: ; represents the weight connection between layers.

[0123] The following explains the multi-task output processing in detail:

[0124] Figure 11Figure 1 is a schematic diagram of a multi-task learning neural network architecture in the embodiments of the present application, see Figure 11 The architecture is divided into three main parts:

[0125] Input layer: input layer, used to receive raw data.

[0126] Common hidden layers: shared hidden layers (such as CNN, LSTM or fully connected layers) extract general features of input data for all tasks.

[0127] Task-specific hidden layers: task-specific layers (including 2 fully connected networks, performing task-related nonlinear transformation on shared features, learning classification decision boundary), specifically processing different tasks. Task1 is a multi-label classification task (Multi-label Classification), and Task2 is a classification task (Classification). Each task has its own specific layer network structure. The arrow indicates the forward propagation direction of feature extraction. The dark gray dot represents the activated output node.

[0128] After flattening the LSTM output, the fully connected layer (64 neurons, Dropout=0.5) performs nonlinear mapping. Then, two output heads are designed, one to complete the binary classification task (the goal of the binary classification task is to divide the input data into two mutually exclusive categories "positive class" and "negative class"), and the output layer uses the sigmoid activation function to output the binary classification probability (leak / non-leak), and the other to complete the multi-classification task, using the Softmax activation function to output the multi-classification probability (leakage level). These two output heads learn general features through shared hidden layer model parameters, reduce overfitting, improve model generalization ability and training efficiency, and reduce computational complexity.

[0129] Step three, develop an engineering early warning system for leakage situations and provide real-time alarm reminders. The built-in soundprint feature library contains a variety of typical leakage soundprints, as well as standardized repair solutions, achieving full-process closed-loop management from hazard identification to intelligent disposal:

[0130] Micro-leakage (Class I): Fault characteristics: weak leakage vibration disturbance, sound wave shows low-frequency energy; Cause analysis: aging of interface rubber sealing element (change in Shore hardness > 20%), pipeline weld micro-porosity (<0.5 mm²), etc.; Recommended treatment scheme: use non-cured rubber asphalt grouting plugging (permeability <1x10 -6 cm / s).

[0131] Medium leakage (level II): fault characteristics: medium leakage vibration disturbance, sound wave shows medium frequency band energy; cause analysis: pipeline corrosion pit depth > 1mm, etc.; recommended processing scheme: injection of polyurethane elastomer seal (curing time 4-6h, elastic modulus ≥10MPa).

[0132] Severe leakage (level III): fault characteristics: severe leakage vibration disturbance, sound wave shows high frequency band energy, and sudden peak value; cause analysis: pipe body rupture (crack width > 2mm), etc.; recommended processing scheme: trigger emergency valve closing protocol, close upstream and downstream butterfly valves, and repair through prefabricated pipe segment.

[0133] The optical fiber vibration auscultation system in the embodiment of the present application is described from the perspective of hardware processing below. Referring to Figure 12 , Figure 12 is a schematic structural diagram of an entity device of the optical fiber vibration auscultation system in the embodiment of the present application.

[0134] It should be noted that Figure 12 The structure of the optical fiber vibration auscultation system shown is only an example, and should not bring any limitation to the function and use range of the embodiment of the present application.

[0135] As Figure 12 shown, the optical fiber vibration auscultation system includes a central processing unit (CPU) 1201, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 1202 or loaded from a storage portion 1208 to a random access memory (RAM) 1203, such as performing the method described in the above embodiment. In the RAM 1203, various programs and data required for system operation are also stored. The CPU 1201, the ROM 1202, and the RAM 1203 are connected to each other through a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0136] The following components are connected to the I / O interface 1205: an input section 1206 including an audio input device, a push button switch, and the like; an output section 1207 including a Liquid Crystal Display (LCD), and an audio output device, a lamp, and the like; a storage section 1208 including a hard disk and the like; and a communication section 1209 including a network interface card such as a LAN (Local Area Network) card, a modem, and the like. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the I / O interface 1205 as necessary. A removable medium 1211 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is attached to the drive 1210 as necessary, so that a computer program read out therefrom is installed in the storage section 1208 as necessary.

[0137] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present application. For example, an embodiment of the present application includes a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing a computer program for executing the method shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network by the communication section 1209, and / or installed from the removable medium 1211. When the computer program is executed by the central processing unit (CPU) 1201, various functions defined in the present application are executed.

[0138] Note that specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any appropriate combination thereof. In the present application, the computer-readable storage medium can be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device.

[0139] The flowcharts and block diagrams in the drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.

[0140] In particular, the optical fiber vibration auscultation system of the embodiment includes a processor and a memory, and the memory stores a computer program. When the computer program is executed by the processor, the optical fiber vibration auscultation method for water supply network leakage provided in the above embodiment is implemented.

[0141] As another aspect, the present application also provides a computer readable storage medium. The storage medium can be included in the optical fiber vibration auscultation system described in the above embodiments, or can exist independently without being assembled into the optical fiber vibration auscultation system. The storage medium carries one or more computer programs. When the one or more computer programs are executed by a processor of the optical fiber vibration auscultation system, the optical fiber vibration auscultation system implements the optical fiber vibration auscultation method for water supply network leakage provided in the above embodiments.

[0142] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0143] Those skilled in the art can understand that all or part of the flow of the above-mentioned embodiment method can be instructed by a computer program to relevant hardware to complete, the program can be stored in a computer readable storage medium, and the program can include the flow of each method embodiment when executed. The foregoing storage medium includes ROM or random storage memory RAM, magnetic disc or optical disc, and various program code storage media.

Claims

1. A method of optical fiber vibration auscultation of water supply network leaks, characterized in that, The method comprises the following steps: Synchronously sampling the double-channel optical fiber vibration signals by using the FPGA master clock to obtain time-aligned pipeline vibration signals and environmental noise signals; Fusing multi-dimensional features of the pipeline vibration signals and the environmental noise signals to obtain a pipeline leakage feature vector; Inputting the pipeline leakage feature vector into a CNN-LSTM hybrid model to obtain pipeline leakage information output by the CNN-LSTM hybrid model after the CNN-LSTM hybrid model performs feature analysis and prediction processing on the pipeline leakage feature vector; The step of fusing multi-dimensional features of the pipeline vibration signals and the environmental noise signals to obtain a pipeline leakage feature vector comprises the following steps: Performing optical time domain reflection processing on the pipeline vibration signals to obtain a phase delay variation of back Rayleigh scattering light; Generating a local oscillation light by using an ultra-narrow linewidth pulsed laser, and superimposing the back Rayleigh scattering light and the local oscillation light on a photodetector to obtain an interference light intensity signal; Filtering the interference light intensity signal and the environmental noise signal by using an improved LMS adaptive filtering algorithm to obtain a leakage vibration signal; Determining a fiber strain feature of a first fiber channel in the double-channel optical fiber vibration signal according to the phase delay variation, wherein the first fiber channel is tightly coupled with a water supply pipeline and is used for collecting the pipeline vibration signals, and a second fiber channel in the double-channel optical fiber vibration signal maintains a preset distance from the water supply pipeline and is used for collecting the environmental noise signals; Extracting time domain features of the leakage vibration signal to obtain a leakage impact feature, and extracting frequency domain features of the leakage vibration signal to obtain a leakage voiceprint feature; Fusing the fiber strain feature, the leakage impact feature and the leakage voiceprint feature to obtain the pipeline leakage feature vector; After the CNN-LSTM hybrid model performs multi-classification processing on the second task feature by using a Softmax activation function to output a pipeline leakage level, the method further comprises the following steps: Extracting a standard leakage voiceprint feature corresponding to the pipeline leakage level from a preset leakage voiceprint feature library; Matching the leakage voiceprint feature in the pipeline leakage feature vector with the standard leakage voiceprint feature to obtain a pipeline leakage type; In a case where it is determined that the pipeline leakage type is a micro-leakage type, determining a low-frequency band energy proportion in the leakage voiceprint feature to obtain a low-frequency band energy proportion result; In a case where it is determined according to the low-frequency band energy proportion result that the low-frequency band energy proportion is greater than a first preset threshold, determining that a pipeline leakage cause is aging of an interface rubber sealing element; In a case where it is determined according to the low-frequency band energy proportion result that the low-frequency band energy proportion is less than or equal to the first preset threshold, determining that the pipeline leakage cause is a pipeline weld micro-pore; In a case where it is determined that the pipeline leakage type is a medium-leakage type, determining a medium-frequency band energy distribution in the leakage voiceprint feature to obtain a medium-frequency band energy distribution result; In a case where it is determined according to the mid-frequency band energy distribution result that a standard deviation of the mid-frequency band energy distribution is greater than a second preset threshold, the leakage cause is determined to be a pipeline corrosion pit; In a case where it is determined that the pipeline leakage type is a serious leakage type, it is detected whether a high-frequency burst peak value exists in the leakage soundprint feature; In a case where it is detected that the high-frequency burst peak value exists in the leakage soundprint feature and an amplitude of the high-frequency burst peak value is greater than a third preset threshold, the leakage cause is determined to be a pipeline body rupture; According to the pipeline leakage cause, fault propagation prediction is performed to obtain a leakage expansion risk level; According to the pipeline leakage level and the leakage expansion risk level, a priority-ordered repair strategy set is generated.

2. The method of claim 1, wherein, The pipeline leakage feature vector is input into the CNN-LSTM hybrid model to obtain pipeline leakage information output by the CNN-LSTM hybrid model after the CNN-LSTM hybrid model performs feature analysis and prediction processing on the pipeline leakage feature vector, and the pipeline leakage information specifically includes: The pipeline leakage feature vector is input into the CNN-LSTM hybrid model to enable the CNN-LSTM hybrid model to perform the following operations: The CNN-LSTM hybrid model performs multi-scale time series modeling processing on the pipeline leakage feature vector to obtain an optimized feature vector in a case where it is determined that the pipeline leakage feature vector is received. The CNN-LSTM hybrid model performs multi-task joint prediction processing on the optimized feature vector to obtain the pipeline leakage information.

3. The method of claim 2, wherein, The CNN-LSTM hybrid model performs multi-scale time series modeling processing on the pipeline leakage feature vector to obtain an optimized feature vector in a case where it is determined that the pipeline leakage feature vector is received, and the multi-scale time series modeling processing specifically includes: The CNN-LSTM hybrid model performs multi-scale feature extraction on the pipeline leakage feature vector by using three groups of parallel convolution kernels of different scales to obtain local time-frequency features. The CNN-LSTM hybrid model performs a max-pooling operation on the local time-frequency features to obtain reduced dimension features. The CNN-LSTM hybrid model performs time series modeling on the reduced dimension features by using a two-layer bidirectional LSTM network to obtain time series correlation features, wherein the two-layer bidirectional LSTM network includes a first LSTM layer for forward propagation and a second LSTM layer for backward propagation. The CNN-LSTM hybrid model performs weighting processing on the time series correlation features by using a self-attention layer to obtain weight features. The CNN-LSTM hybrid model performs non-linear mapping on the weight features by using a fully connected layer to obtain the optimized feature vector.

4. The method of claim 2, wherein, The CNN-LSTM hybrid model performs multi-task joint prediction processing on the optimized feature vector to obtain the pipeline leakage information, and the multi-task joint prediction processing specifically includes: The CNN-LSTM hybrid model inputs the optimized feature vector after flattening into a shared fully connected layer to perform first non-linear mapping to obtain shared features. The CNN-LSTM hybrid model inputs the shared features into a first task-specific layer to perform second non-linear transformation to obtain first task features. The CNN-LSTM hybrid model utilizes a Sigmoid activation function to perform binary classification processing on the first task feature to output a pipeline leakage probability, wherein the pipeline leakage information comprises the pipeline leakage probability. The CNN-LSTM hybrid model inputs the shared feature into a second task-specific layer for a third nonlinear transformation to obtain a second task feature. The CNN-LSTM hybrid model utilizes a Softmax activation function to perform multi-classification processing on the second task feature to output a pipeline leakage grade, wherein the pipeline leakage information comprises the pipeline leakage grade.

5. The method of claim 1, wherein, The FPGA master clock is used to synchronously sample the double-channel optical fiber vibration signals to obtain time-aligned pipeline vibration signals and environmental noise signals, specifically including: The FPGA master clock is controlled to send a Sync message to a vibration collection terminal, and a first sending timestamp of the FPGA master clock sending the Sync message is recorded; In a case where it is determined that the vibration collection terminal receives the Sync message, a first receiving timestamp of the vibration collection terminal receiving the Sync message is recorded; The vibration collection terminal is controlled to return a Delay_Req message to the FPGA master clock, and a second sending timestamp of the vibration collection terminal returning the Delay_Req message is recorded; In a case where it is determined that the FPGA master clock receives the Delay_Req message, a second receiving timestamp of the FPGA master clock receiving the Delay_Req message is recorded; A link delay time is calculated according to the first sending timestamp, the first receiving timestamp, the second sending timestamp, and the second receiving timestamp, wherein the link delay time is used to represent signal transmission delay between the FPGA master clock and the vibration collection terminal; According to the link delay time, time compensation processing is performed on initial pipeline vibration signals and initial environmental noise signals, and interpolation alignment processing is performed on the compensated initial pipeline vibration signals and initial environmental noise signals to obtain time-aligned pipeline vibration signals and environmental noise signals.

6. An optical fiber vibration stethoscopic system characterized by, The optical fiber vibration auscultation system comprises one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code comprises computer instructions, and the one or more processors invoke the computer instructions to enable the optical fiber vibration auscultation system to perform the method of any one of claims 1-5.

7. A computer-readable storage medium comprising instructions, characterized in that, When the instructions run on the optical fiber vibration auscultation system, the optical fiber vibration auscultation system performs the method of any one of claims 1-5.

8. A computer program product, characterised in that, When the computer program product runs on the optical fiber vibration auscultation system, the optical fiber vibration auscultation system performs the method of any one of claims 1-5.

Images