Intelligent water data acquisition and control method, system and terminal device

Abstract

The invention provides an intelligent water use data acquisition and control method and system and a terminal device, and discloses the intelligent water use data acquisition and control method and system and the terminal device which are suitable for urban water supply and industrial buried pipeline tiny water leakage scenes. A distributed optical fiber acoustic sensor, a temperature measurement sensor, a nano-film vibration sensor array and a high-precision pressure sensor are arranged to collect multi-physical field data; multi-modal data space-time alignment and feature fusion are achieved through multi-scale wavelet transformation and dynamic time warping, temperature interference is eliminated through a long and short term memory-causal network, and space-time features are processed in combination with a Transform network; the particle swarm optimization algorithm constrained based on the Navier-Stokes equation cooperates with the physical information generative adversarial network, high-precision inversion and positioning of the coordinates of the leakage points are achieved, multi-physics field data and an intelligent algorithm are fused, the positioning precision and reliability of the leakage points are improved, and the method has adaptability to pipelines made of different materials.

Description

Technical Field

[0001] The present invention belongs to the field of smart water technology. Specifically, it relates to an intelligent water use data acquisition and control method, system, and terminal device. It adopts a method, system, and terminal device for detecting and locating tiny leaks in buried pipelines based on multi-physical field fusion perception and driven by intelligent algorithms. It is suitable for urban water supply and industrial use of tiny leaks in buried pipelines. Background Art

[0002] In the field of smart water services, buried pipelines serve as infrastructure for urban water supply and industrial fluid transportation. Their safe operation is directly related to resource efficiency and public safety.

[0003] However, small leaks caused by pipeline aging, corrosion or external force damage are difficult to detect in a timely manner, especially in non-metallic pipelines such as plastic, composite materials or deep buried scenarios, where leak location faces significant technical challenges.

[0004] To this end, the present invention provides an intelligent water use data acquisition and control method, system and terminal device. Summary of the Invention

[0005] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] In a first aspect, the present invention provides an intelligent water use data acquisition and control method, comprising:

[0008] S1: Arrange sensors to collect multi-physics field data, including: arrival time of the first acoustic wave at the leak point, temperature gradient along the pipeline, vibration signal, and pressure data;

[0009] S2: Based on the collected multi-physics field data, a multi-scale wavelet transform is used to perform time-frequency decomposition on the time series of the first arrival of the acoustic wave at the leak point and the vibration signal. This is used to detect mutation points and optimize feature reliability. Dynamic time warping and cross-spectral entropy are then used to align the time axes of different sensors and extract effective feature segments.

[0010] S3: After time axis alignment and effective feature extraction, the multi-physics field data is analyzed through a long short-term memory-causal network to analyze the relationship between temperature gradient data and vibration energy. A physical relationship model is constructed to calculate the temperature influence factor. The temperature influence factor is used to compensate for the vibration energy and eliminate the noise introduced by ambient temperature fluctuations.

[0011] S4: The vibration energy after eliminating temperature interference is combined with the joint feature matrix data in the time and space domains to locate the leakage point through the optimization algorithm constrained by physical equations and the generative adversarial network.

[0012] In a second aspect, the present invention provides an intelligent water use data acquisition and control system, comprising:

[0013] Multi-physics field data acquisition module: Arrange sensors to collect multi-physics field data, including: arrival time of the first acoustic wave at the leak point, temperature gradient along the pipeline, vibration signal, and pressure data;

[0014] Multimodal Data Spatiotemporal Alignment Module: Based on the collected multi-physics field data, the module performs time-frequency decomposition of the acoustic first wave arrival time series and vibration signals including the leak point through multi-scale wavelet transform, detects mutation points to optimize feature reliability, and uses dynamic time warping and cross-spectral entropy to align the time axes of different sensors and extract effective feature segments.

[0015] Temperature Interference Elimination and Feature Fusion Module: This module uses a long-short-term memory-causal network to analyze the relationship between temperature gradient data and vibration energy in a portion of multi-physics field data that has undergone time axis alignment, effective feature optimization, and extraction. This module then constructs a physical relationship model, calculates the temperature influence factor, and uses this temperature influence factor to compensate for vibration energy, eliminating noise introduced by ambient temperature fluctuations.

[0016] Physical constraint intelligent positioning module: The vibration energy after temperature interference is eliminated is combined with the joint feature matrix data in the time and space domains to locate the leakage point through the optimization algorithm constrained by physical equations and the generative adversarial network.

[0017] In a third aspect, the present invention provides an intelligent water use data acquisition and control terminal device, comprising:

[0018] Distributed fiber optic acoustic wave sensors, distributed fiber optic temperature sensors, nano-thin film vibration sensor arrays, high-precision pressure sensors, and computer processing units.

[0019] The beneficial effects of the present invention are as follows:

[0020] 1. Deploy distributed fiber optic acoustic wave sensors, temperature sensors, nano-thin film vibration sensor arrays, and high-precision pressure sensors to synchronously collect multi-dimensional data such as acoustic waves, temperature, vibration, and pressure, covering the key physical characteristics of leak detection. Taking advantage of the low acoustic impedance of plastic pipes, sonar patches are used to enhance signal coupling, address the detection differences between metal and non-metallic pipes, and achieve adaptability to pipes of different materials. Multi-scale wavelet transforms are used to perform time-frequency decomposition of non-stationary signals. Signal mutation points are detected in layers to avoid interference from noise in a single frequency band and improve feature reliability. Dynamic time warping (DTW) and cross spectral entropy (CSE) are used to address timing offsets caused by differences in sampling frequencies of different sensors, ensuring precise alignment of multimodal data in the time and frequency domains, providing a synchronized and reliable data foundation for subsequent analysis.

[0021] 2. A long short-term memory-causal network (LSTM-CN) is used to build a physical relationship model between temperature gradient and vibration energy attenuation, dynamically learn temperature influencing factors, quantitatively compensate for the interference of temperature fluctuations on vibration signals, and ensure that the detection results are not affected by ambient temperature changes; combined with the Transformer network and attention mechanism, multimodal data is jointly extracted in the spatiotemporal domain: the time stream Transformer processes the time series of sound wave arrival and identifies multipath reflection interference; the spatial stream Transformer integrates the GIS pipeline network topology, weights high leakage risk nodes, and guides the algorithm to prioritize searching high-risk areas to improve detection efficiency; the particle swarm optimization (PSO) algorithm based on the Navier-Stokes equation constraints introduces pipeline material parameters and fluid mechanics equations to dynamically adjust the particle search step size to ensure that the search path conforms to physical laws; the physical information generative adversarial network (PI-GAN) filters candidate solutions that violate the wave equation in real time and eliminates physically infeasible solutions; the collaborative positioning algorithm achieves high-precision inversion of the leak point coordinates, with a positioning error of ≤0.5 meters, and the physical consistency is verified through COMSOL simulation to ensure the reliability of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The present invention will be further described below with reference to the accompanying drawings.

[0023] Figure 1 is a flowchart of the steps of Example 1 of the present invention;

[0024] Figure 2 is a system module diagram of embodiment 2 of the present invention;

[0025] Figure 3 This is a device connection diagram of Example 3 of the present invention. DETAILED DESCRIPTION

[0026] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0027] Example 1

[0028] like Figure 1 As shown, an intelligent water use data acquisition and control method according to an embodiment of the present invention includes:

[0029] S1: Arrange sensors to collect multi-physics field data, including: arrival time of the first acoustic wave at the leak point, temperature gradient along the pipeline, vibration signal, and pressure data;

[0030] By arranging distributed fiber optic acoustic wave sensors DAS, distributed fiber optic temperature sensors DTS, nano-thin film vibration sensor arrays, and high-precision pressure sensors, they are used to collect multi-physical field data;

[0031] In some embodiments, distributed fiber optic acoustic wave sensors (DAS) are deployed at preset distances along the pipeline to extract the arrival time of the first wave of the sound wave at the leak point. Distributed fiber optic temperature sensors (DTS) are also integrated to monitor the temperature gradient along the pipeline. Given the low acoustic impedance of plastic pipelines, sonar patches are used to improve the transmission efficiency of leakage sound to the sensor and enhance the coupling of leakage sound signals. It should be noted that distributed fiber optic acoustic wave sensors (DAS), distributed fiber optic temperature sensors (DTS), and nano-thin film vibration sensor arrays can be directly deployed on metal pipelines.

[0032] In some embodiments, a nano-film vibration sensor array is deployed on the outer wall of the pipeline to monitor the vibration signal generated by the leak point, and high-precision pressure sensors are installed at preset distances along the pipeline to collect pressure data in real time, and a computer processing unit is used to calculate the real-time pressure derivative. Unit: Pa / s;

[0033] The arrival time of the first wave of the acoustic wave at the leak point, the temperature gradient along the pipeline, the vibration signal, and the pressure data constitute multi-physical field data;

[0034] S2: Based on the collected multi-physics field data, a multi-scale wavelet transform is used to perform time-frequency decomposition on the time series of the first arrival of the acoustic wave at the leak point and the vibration signal. This is used to detect mutation points and optimize feature reliability. Dynamic time warping and cross-spectral entropy are then used to align the time axes of different sensors and extract effective feature segments.

[0035] The acoustic and vibration signals generated by buried pipeline leaks are non-stationary signals, and direct analysis is difficult to capture detailed features.

[0036] In some embodiments, the distributed fiber acoustic sensor (DAS) and nano-thin film vibration sensor array data are subjected to wavelet packet decomposition to generate three layers of time-frequency sub-bands, including a low-frequency layer, a mid-frequency layer, and a high-frequency layer; a Daubechies wavelet basis function is used with a decomposition level of J=3 to achieve multi-resolution analysis of non-stationary signals;

[0037] Detect signal mutation points in each sub-band, calculate the maximum value of the wavelet coefficient modulus, and locate the signal mutation points. Mutation points are direct evidence of the existence of leakage points. The mutation points include: pressure steps and vibration peaks. Mutation points are detected independently in the low-frequency layer, mid-frequency layer, and high-frequency layer to avoid interference from noise in a single frequency band.

[0038] Through the cross-subband mutation point correlation analysis, the false features of a single frequency band are eliminated;

[0039] Layered detection can avoid single frequency band noise interference to optimize feature reliability;

[0040] The sampling frequencies of the distributed fiber acoustic sensor (DAS) and the nanofilm vibration sensor array may be different, resulting in a mismatch in the timestamps of the same physical event in the data of the two sensors;

[0041] In some embodiments, dynamic time warping (DTW) is used to stretch or compress the time axis of the feature point sequence of the distributed optical fiber acoustic wave sensor (DAS) and the nano-film vibration sensor array. The Sakoe-Chiba bandwidth constraint is used to minimize the matching distance, limit the DTW search range, avoid global search, reduce computational difficulty, and improve real-time performance. The window size r is set to 0.1N, where N is the sequence length. Turning points are selected on the DTW optimal path as time synchronization markers.

[0042] Calculate the cross power spectrum density (CPSD) of the 200ms data segment before and after the synchronization mark point. The calculation formula is: Among them, A(f) and B(f) are the results of windowed fast Fourier transform, K is the number of segments, and B k * (f) represents the complex conjugate of B(f), A is the signal data of the distributed optical fiber acoustic sensor DAS, B is the signal data of the nano-film vibration sensor array;

[0043] By defining the cross spectral entropy CSE, the similarity of two signals in the frequency domain is reflected, specifically: CSExy = -Σ f P xy (f)logP xy (f), where The calculated cross spectral entropy CSE value is analyzed. If the cross spectral entropy CSE value is greater than 0.85, it is determined to be a valid synchronization feature segment and the valid feature segment is extracted;

[0044] After wavelet packet decomposition and dynamic time warping (DTW) alignment, the data from the distributed fiber acoustic sensor (DAS) and the nanofilm vibration sensor array are synchronized on the time axis and their frequency components are separated. This ensures that the subsequent cross-spectral entropy analysis (CSE) can accurately calculate the similarity of the two signals in the frequency domain, avoiding misjudgments caused by timing or frequency mismatches.

[0045] S3: After time axis alignment and effective feature extraction, the multi-physics field data is analyzed through a long short-term memory-causal network to analyze the relationship between temperature gradient data and vibration energy. A physical relationship model is constructed to calculate the temperature influence factor. The temperature influence factor is used to compensate for the vibration energy and eliminate the noise introduced by ambient temperature fluctuations.

[0046] First, a long short-term memory-causal network (LSTM-CN) is used to eliminate temperature interference and build a physical relationship model between temperature gradient and vibration energy attenuation. A Transformer network is then used to process spatiotemporal features, ultimately providing high-quality input for the physical constraint location model and improving leak location accuracy.

[0047] Perform sliding window processing on the data sequence collected by the distributed optical fiber temperature sensor DTS, calculate the temperature gradient, and calculate multiple temperature gradients to obtain a temperature gradient sequence;

[0048] The vibration signal data collected by the distributed optical fiber acoustic sensor (DAS) is subjected to wavelet denoising to extract the vibration energy characteristic sequence.

[0049] The temperature gradient sequence and vibration energy feature sequence are input into the input layer of the long short-term memory network (LSTM). The temporal relationship between temperature and vibration is captured through the bidirectional LSTM unit in the hidden layer. The output layer outputs the temperature influence factor α, which represents the weight of the impact of temperature change on the vibration signal.

[0050] In the causal network CN, a causal graph G = (V, E) of temperature gradient and vibration energy is constructed, where node V = {T g ,H},H vibration energy, represents the energy characteristics related to vibration, T g is the temperature gradient, reflecting the change of temperature, and E represents the strength of causal relationship. The causal structure is learned by PC algorithm, and the causal strength is calculated using conditional mutual information CMI: CMI(T g ;E v |T g-1 )=H(T g ;E v |T g-1 )-H(T g |T g-1 )-H(E v |T g-1 ), where H(·) is the information entropy, T g is the temperature gradient, E v is the vibration energy of H, T g-1 is the temperature gradient at the previous moment, when CMI(T g ;E v |T g-1 )>0.3, it indicates that the temperature gradient change has a substantial impact on the vibration energy;

[0051] Calculate the vibration energy after temperature compensation using the formula: H′=H·(1-α t ·|T g |) to eliminate the interference of temperature fluctuation on the vibration signal, where α tis the temperature influence factor, which is obtained through dynamic learning of the LSTM network, H is the original vibration energy, T g The temperature gradient represents the amount of change in temperature and is taken in absolute value to focus on the magnitude of the temperature change rather than the direction;

[0052] Perform feature enhancement on the compensated vibration signal: Among them, σ T is the standard deviation of the temperature gradient, used to calculate the temperature gradient |T g |Perform normalization processing to reasonably control the amplitude of feature enhancement and highlight the vibration characteristics of the temperature abnormality area. H′ is the vibration energy after temperature compensation. T g is the temperature gradient;

[0053] Traditional positioning algorithms do not fully utilize the topology of the pipe network, wasting computing resources in areas with high leakage risk.

[0054] In some embodiments, a temporal stream transformer is used to process the TDOA acoustic wave arrival time series, and a self-attention mechanism is used to identify multipath reflection interference. A spatial stream transformer is used to encode the GIS pipe network topology into a graph structure, and high leakage risk nodes are weighted through the GAT network, so that the auxiliary positioning algorithm prioritizes the search for high-risk areas.

[0055] The calibrated multimodal data is filtered through the attention mechanism and the spatiotemporal joint feature matrix is ​​output as the input of the downstream positioning model.

[0056] The attention mechanism is used to perform weighted screening on the calibrated multimodal data acoustic wave features, temperature gradient sequence, vibration energy feature sequence, and real-time pressure derivative.

[0057] When the temperature gradient sequence shows anomalies, the weight of the vibration energy feature is reduced;

[0058] When a step occurs in the pressure derivative, the weight of the acoustic wave feature is increased. The pressure step directly reflects the transient pressure wave generated by the leak and is strongly correlated with the acoustic wave feature.

[0059] Dynamically adjust the importance of features based on real-time data quality to avoid single-modality data anomalies dominating the positioning results;

[0060] The filtered features are integrated into a joint feature matrix in the time and space domains, where the rows represent time series, the columns represent spatial nodes, and each element contains multi-physics field features;

[0061] S4: The vibration energy characteristics, acoustic wave characteristics and the joint feature matrix in the time and space domain after eliminating temperature interference are used to locate the leakage point through the optimization algorithm constrained by physical equations and the generative adversarial network.

[0062] The PSO algorithm, based on Navier-Stokes equation constraints, introduces the pipeline fluid dynamic equation during the particle swarm optimization process. Based on the real-time pressure derivative and material parameters, it calculates the correction coefficient of the pressure wave propagation velocity caused by the leak and dynamically adjusts the particle search step size. The adjoint method automatically generates the gradient direction, so that the particle movement path conforms to the mass conservation equation and reduces invalid iterations.

[0063] Specifically, each particle is defined to represent the coordinates (x, y, z) of a potential leak point, where the dimensions correspond to the spatial coordinate system of the pipe network;

[0064] Physical parameter input, including real-time pressure derivative Pipe material parameters and fluid properties; fluid properties include: water density and water viscosity; pipe material parameters include: material type, elastic modulus F (unit: GPa), wall thickness h (unit: mm), inner diameter D (unit: mm);

[0065] Based on the propagation velocity modeling of the Navier-Stokes equation, the pressure wave velocity c generated by the leakage satisfies the simplified wave equation: Where K is the bulk modulus of the fluid, D is the inner diameter of the pipe, h is the pipe wall thickness, h pipe wall thickness refers to the thickness of the pipe wall, and F is the elastic modulus of the pipe material, which measures the ability of the pipe material to resist elastic deformation;

[0066] It should be noted that the elastic modulus F is strongly correlated with the pipe material. The value of metal pipes is usually 100-300 times that of plastic pipes, which directly affects the calculation results of the pressure wave velocity c.

[0067] According to the real-time pressure derivative Calculate the correction factor λ with the material parameters: Where η is the correction factor used to dynamically adjust the wave velocity estimate;

[0068] Particle search step update, particle speed v i and position x i Update formula: Where ω is the inertia weight, is the velocity of particle i at time t+1, c1 and c2 are learning factors, r1 and r2 are random numbers between 0 and 1, which increase the randomness of the search process and avoid the algorithm falling into the local optimum, λ is the correction coefficient, and pbest i The individual optimal position of particle i is the best position found during the particle's own search process, and the gbest global optimal position is the best position found during the search process of all particles. The position of particle i at time t+1 is determined by the current position and speed Iterative update is obtained;

[0069] By introducing physical wave velocity constraints, the particle step size is made consistent with the calculation results of the actual leakage propagation law;

[0070] The objective function f(x) is constructed by automatic derivation through the adjoint method to convert the multi-sensor TDOA positioning error into the gradient Δf(x) through the adjoint method, guiding the particles to move in the direction with the fastest error decrease, thus reducing blind search: Among them, α is the learning rate, The objective function f(x) is The gradient at , guides the particle to move in the direction where the positioning error decreases fastest;

[0071] The mass conservation equation constrains the particle movement path to satisfy: The particle velocity is projected onto the divergence-free subspace through the projection method to ensure that the search path conforms to the physical laws and avoids generating invalid solutions that violate fluid mechanics.

[0072] Data input and physical criterion definition, input layer input pipeline network parameter vector: Contains pressure p, which is the raw pressure data collected in real time by a high-precision pressure sensor, and temperature gradient T g The temperature gradient along the pipeline is monitored by the distributed optical fiber temperature sensor DTS, and E is the pipeline material parameter;

[0073] Generator and discriminator architecture, input network parameters θ plus random noise z; output mixed waveform s gen = G(θ,z), which contains the simulated leak signal and noise. The pipeline network parameters include: internal pressure data collected in real time by high-precision pressure sensors, temperature gradient data along the pipeline monitored by distributed fiber optic temperature sensors (DTS), and relevant characteristic parameters of the pipeline material;

[0074] Physical constraints are embedded, and a wave equation penalty term is added to the generator loss function: Among them, E θ,z For the expectation operation on θ and z, For mixed waveforms gen The partial derivative of time t reflects the rate of change of the waveform over time, c is the pressure wave velocity generated by the leak, For mixed waveforms gen The Laplace operator describes the second-order change of the waveform in space. The input s gen , output the true or false probability D(s);

[0075] Adversarial training and interference filtering, training objectives: Among them, λ phys is the physical constraint weight, balancing data authenticity and physical rationality, min GOptimize the generator G so that the waveform it generates can deceive the discriminator as much as possible, max D Optimize the discriminator D to make it more accurate in distinguishing between real and fake waveforms, E θ,z Again perform the expectation operation on θ and z;

[0076] Interference filtering mechanism: After training, PI-GAN's discriminator can identify interference signals that do not conform to physical laws, thereby improving the signal purity of the input positioning model;

[0077] For example, when D(s) < 0.6, it is determined to be an interference signal;

[0078] Multimodal feature input and leak point coordinate inversion: The input is the spatiotemporal joint feature matrix data processed by LSTM-CN and Transformer. The dimension is: T×N×M, where T is the time step, N is the number of sensor nodes used to collect multimodal data, and M is the number of physical field modes. The spatiotemporal joint feature matrix data includes: TDOA sequence, vibration energy compensation value, GIS topology encoded node risk weight, and sensor location coordinates.

[0079] PSO and PI-GAN collaborative positioning steps:

[0080] Initialization search: PSO particles are randomly distributed in the pipe network space, and the matching degree between the initial feature matrix and the particle position, as well as the error between the TDOA time difference and the theoretical propagation time, are calculated;

[0081] PI-GAN pre-screening: For high-error particles, PI-GAN is used to generate a theoretical leakage waveform at that location, which is then compared with the actual characteristic matrix to eliminate physically infeasible solutions.

[0082] Further iterative optimization is performed: PSO updates the particle positions according to the Navier-Stokes constraints, and PI-GAN filters candidate solutions that violate the wave equation in real time;

[0083] Convergence judgment: When the positioning error of the particle swarm optimal solution is less than the threshold, such as 0.5 meters or the number of iterations reaches the upper limit, the coordinates of the leakage point (x * ,y * ,z * );

[0084] COMSOL was used to simulate the leakage scenario at this coordinate. The original pressure data and the temperature gradient distribution along the pipeline were calculated. The sensor position coordinates were compared with the multimodal feature matrix actually acquired to verify the physical consistency of the positioning results: Among them, f sim is the simulation feature, f real,m is the measured characteristic;

[0085] The sensor position coordinates are: the specific coordinate values ​​of the distributed optical fiber acoustic sensor DAS, distributed optical fiber temperature sensor DTS, nano-thin film vibration sensor array, and high-precision pressure sensor in the pipe network;

[0086] If the verification index is lower than the preset threshold, the leak location is determined to be valid;

[0087] The technical solution of the embodiment of the present invention is as follows: deploying distributed fiber-optic acoustic wave (DAS), temperature sensors (DTS), nano-thin film vibration sensor arrays, and high-precision pressure sensors to collect acoustic wave, temperature, vibration, and pressure data; attaching sonar patches to plastic pipes to enhance signal coupling; and performing data processing and feature fusion: detecting signal mutation points through wavelet packet decomposition, aligning time series differences using dynamic time warping (DTW) and cross-spectral entropy (CSE); eliminating temperature interference with the help of a long short-term memory-causal network (LSTM-CN), and generating a joint feature matrix in the spatiotemporal domain by combining a Transformer network and an attention mechanism.

[0088] The particle swarm optimization (PSO) algorithm based on Navier-Stokes equation constraints works in conjunction with the physical information generative adversarial network (PI-GAN). Through pressure wave velocity modeling, adjoint method gradient guidance, and wave equation interference filtering, high-precision inversion of leak point coordinates is achieved, with a positioning error of ≤0.5 meters. The physical consistency is verified through COMSOL simulation. This solution integrates multimodal data and intelligent algorithms to improve detection accuracy and engineering adaptability.

[0089] Example 2

[0090] like Figure 2 As shown, based on Example 1, the present invention provides an intelligent water use data acquisition and control system, including the following modules:

[0091] Multi-physics field data acquisition module: Arrange sensors to collect multi-physics field data, including: arrival time of the first acoustic wave at the leak point, temperature gradient along the pipeline, vibration signal, and pressure data;

[0092] Distributed fiber optic acoustic sensors (DAS) are deployed at preset distances along the pipeline to extract the arrival time of the first acoustic wave at the leak point. Distributed fiber optic temperature sensors (DTS) are also integrated to monitor the temperature gradient along the pipeline. Given the low acoustic impedance of plastic pipelines, sonar patches are used to improve the transmission efficiency of leak sound to the sensors and enhance the coupling of leak sound signals. It should be noted that distributed fiber optic acoustic sensors (DAS), distributed fiber optic temperature sensors (DTS), and nano-thin film vibration sensor arrays can be directly deployed on metal pipelines.

[0093] A nano-film vibration sensor array is deployed on the outer wall of the pipeline to monitor the vibration signal generated by the leak point. High-precision pressure sensors are installed at preset distances along the pipeline to collect pressure data in real time and calculate the real-time pressure derivative. Unit: Pa / s;

[0094] Multimodal Data Spatiotemporal Alignment Module: Based on the collected multi-physics field data, the module performs time-frequency decomposition of the acoustic first wave arrival time series and vibration signals including the leak point through multi-scale wavelet transform, detects mutation points to optimize feature reliability, and uses dynamic time warping and cross-spectral entropy to align the time axes of different sensors and extract effective feature segments.

[0095] The distributed fiber acoustic sensor (DAS) and nano-film vibration sensor array data are decomposed into wavelet packets, the signal mutation points are detected in each sub-band, the maximum value of the wavelet coefficient modulus is calculated, the signal mutation points are located, and the feature reliability is optimized.

[0096] Dynamic time warping (DTW) is used to stretch or compress the time axis of the feature point sequences of the distributed fiber acoustic sensor (DAS) and the nanofilm vibration sensor array. The Sakoe-Chiba bandwidth constraint is used to minimize the matching distance, limit the DTW search range, avoid global search, reduce computational difficulty, and improve real-time performance. The cross-power spectral density (CPSD) of the data segment 200ms before and after the synchronization mark point is calculated. The cross-spectral entropy (CSE) is defined to reflect the similarity of the two signals in the frequency domain. If the CSE value is greater than 0.85, it is determined to be a valid synchronization feature segment and extracted.

[0097] Temperature Interference Elimination and Feature Fusion Module: This module uses a long-short-term memory-causal network to analyze the relationship between temperature gradient data and vibration energy in a portion of multi-physics field data after time axis alignment and effective feature extraction. This module then constructs a physical relationship model, calculates the temperature influence factor, and uses this temperature influence factor to compensate for vibration energy, eliminating noise introduced by ambient temperature fluctuations.

[0098] The Long Short-Term Memory Causal Network (LSTM-CN) eliminates temperature interference and builds a physical relationship model between temperature gradient and vibration energy attenuation. The Transformer network then processes spatiotemporal features, ultimately providing high-quality input for the physical constraint location model and improving leak location accuracy.

[0099] Perform sliding window processing on the data sequence collected by the distributed optical fiber temperature sensor DTS, calculate the temperature gradient, and calculate multiple temperature gradients to obtain a temperature gradient sequence;

[0100] The vibration signal data collected by the distributed optical fiber acoustic sensor (DAS) is subjected to wavelet denoising to extract the vibration energy characteristic sequence.

[0101] The temperature gradient sequence and vibration energy feature sequence are input into the input layer of the long short-term memory network (LSTM). The temporal relationship between temperature and vibration is captured through the bidirectional LSTM unit in the hidden layer. The output layer outputs the temperature influence factor α, which represents the weight of the impact of temperature change on the vibration signal.

[0102] In the causal network CN, a causal graph of temperature gradient and vibration energy is constructed. The causal structure is learned through the PC algorithm, and the causal strength is calculated using conditional mutual information (CMI). When the causal strength is greater than the threshold, it indicates that the temperature gradient change has a substantial impact on the vibration energy.

[0103] Calculate the vibration energy after temperature compensation using the formula: H′=H·(1-α t ·|T g |) to eliminate the interference of temperature fluctuation on the vibration signal, where α t is the temperature influence factor, which is obtained through dynamic learning of the LSTM network, H is the original vibration energy, T g The temperature gradient represents the amount of change in temperature and is taken in absolute value to focus on the magnitude of the temperature change rather than the direction;

[0104] Perform feature enhancement on the compensated vibration signal: Among them, σ T is the standard deviation of the temperature gradient, used to calculate the temperature gradient |T g |Perform normalization processing to reasonably control the amplitude of feature enhancement and highlight the vibration characteristics of the temperature abnormality area. H′ is the vibration energy after temperature compensation. T g is the temperature gradient;

[0105] Traditional positioning algorithms do not fully utilize the topology of the pipe network, wasting computing resources in areas with high leakage risk.

[0106] The temporal stream transformer is used to process the arrival time series of TDOA sound waves, and the self-attention mechanism is used to identify multipath reflection interference. The spatial stream transformer is used to encode the GIS pipe network topology into a graph structure, and the high leakage risk nodes are weighted through the GAT network. The auxiliary positioning algorithm prioritizes the search for high-risk areas.

[0107] The calibrated multimodal data is filtered through the attention mechanism and the spatiotemporal joint feature matrix is ​​output as the input of the downstream positioning model.

[0108] The attention mechanism is used to perform weighted screening on the calibrated multimodal data acoustic wave features, temperature gradient sequence, vibration energy feature sequence, and real-time pressure derivative.

[0109] When the temperature gradient sequence shows anomalies, the weight of the vibration energy feature is reduced;

[0110] When a step occurs in the pressure derivative, the weight of the acoustic wave feature is increased. The pressure step directly reflects the transient pressure wave generated by the leak and is strongly correlated with the acoustic wave feature.

[0111] Dynamically adjust the importance of features based on real-time data quality to avoid single-modality data anomalies dominating the positioning results;

[0112] The filtered features are integrated into a joint feature matrix in the time and space domains, where the rows represent time series, the columns represent spatial nodes, and each element contains multi-physics field features;

[0113] Physical Constraint Intelligent Positioning Module: This module uses a physical equation-constrained optimization algorithm and a generative adversarial network to coordinate the vibration energy characteristics, acoustic wave characteristics, and the joint feature matrix in the time and space domains after eliminating temperature interference to locate leaks.

[0114] The PSO algorithm, based on Navier-Stokes equation constraints, introduces the pipeline fluid dynamic equation during the particle swarm optimization process. Based on the real-time pressure derivative and material parameters, it calculates the correction coefficient of the pressure wave propagation velocity caused by the leak and dynamically adjusts the particle search step size. The adjoint method automatically generates the gradient direction, so that the particle movement path conforms to the mass conservation equation and reduces invalid iterations.

[0115] Data input and physical criterion definition, input network parameter vector into the input layer, generator and discriminator architecture, input network parameters plus random noise, output mixed waveform, add wave equation penalty term to generator loss function, input mixed waveform, output true or false probability;

[0116] Adversarial training and interference filtering, training objectives: Among them, λ phys is the physical constraint weight, balancing data authenticity and physical rationality, min G Optimize the generator G so that the waveform it generates can deceive the discriminator as much as possible, max D Optimize the discriminator D to make it more accurate in distinguishing between real and fake waveforms, E θ,z Again perform the expectation operation on θ and z;

[0117] PSO and PI-GAN collaborative positioning steps:

[0118] Initialization search: PSO particles are randomly distributed in the pipe network space, and the matching degree between the initial feature matrix and the particle position, as well as the error between the TDOA time difference and the theoretical propagation time, are calculated;

[0119] PI-GAN pre-screening: For high-error particles, PI-GAN is used to generate a theoretical leakage waveform at that location, which is then compared with the actual characteristic matrix to eliminate physically infeasible solutions.

[0120] Further iterative optimization is performed: PSO updates the particle positions according to the Navier-Stokes constraints, and PI-GAN filters candidate solutions that violate the wave equation in real time;

[0121] Convergence judgment: When the positioning error of the particle swarm optimal solution is less than the threshold, such as 0.5 meters or the number of iterations reaches the upper limit, the coordinates of the leakage point (x * ,y * ,z * );

[0122] COMSOL was used to simulate the leakage scenario at this coordinate. The original pressure data and the temperature gradient distribution along the pipeline were calculated. The sensor position coordinates were compared with the multimodal feature matrix actually acquired to verify the physical consistency of the positioning results: Among them, f sim is the simulation feature, f real,m is the measured characteristic;

[0123] The sensor position coordinates are: the specific coordinate values ​​of the distributed optical fiber acoustic sensor DAS, distributed optical fiber temperature sensor DTS, nano-thin film vibration sensor array, and high-precision pressure sensor in the pipe network;

[0124] If the verification index is lower than the preset threshold, the leak location is determined to be valid;

[0125] Example 3

[0126] like Figure 3 As shown, based on Example 1 and Example 2, the present invention provides an intelligent water use data acquisition and control terminal device, including:

[0127] Distributed fiber optic acoustic wave sensors, distributed fiber optic temperature sensors, nano-thin film vibration sensor arrays, high-precision pressure sensors, and computer processing units;

[0128] The distributed optical fiber acoustic wave sensor is used to collect the arrival time of the first acoustic wave at the leakage point;

[0129] The distributed optical fiber temperature sensor is used to monitor the temperature gradient along the pipeline;

[0130] The nano-film vibration sensor array is used to monitor the vibration signal generated by the leak point;

[0131] The high-precision pressure sensor is used to collect pressure data;

[0132] The computer processing unit uses a computer program to process the data collected by the distributed optical fiber acoustic wave sensor, distributed optical fiber temperature sensor, nano-thin film vibration sensor array, and high-precision pressure sensor. At the same time, the computer program can execute the calculation process required by any step or module in the intelligent water use data acquisition and control method and system described in Examples 1 and 2 of the present invention;

[0133] The distributed optical fiber acoustic wave sensor, distributed optical fiber temperature sensor, nano film vibration sensor array, high-precision pressure sensor and computer processing unit are connected via a CAN bus;

[0134] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent water use data acquisition and control method, characterized by: include: S1: Arrange sensors to collect multi-physics field data, including: arrival time of the first acoustic wave at the leak point, temperature gradient along the pipeline, vibration signal, and pressure data; S2: Based on the collected multi-physics field data, a multi-scale wavelet transform is used to perform time-frequency decomposition on the time series of the first arrival of the acoustic wave at the leak point and the vibration signal. This is used to detect mutation points and optimize feature reliability. Dynamic time warping and cross-spectral entropy are then used to align the time axes of different sensors and extract effective feature segments. S3: After time axis alignment and effective feature extraction, the multi-physics field data is analyzed through a long short-term memory-causal network to analyze the relationship between temperature gradient data and vibration energy. A physical relationship model is constructed to calculate the temperature influence factor. The temperature influence factor is used to compensate for the vibration energy and eliminate the noise introduced by ambient temperature fluctuations. S4: The vibration energy after eliminating temperature interference is combined with the joint feature matrix data in the time and space domains to locate the leakage point through the optimization algorithm constrained by physical equations and the generative adversarial network.

2. The intelligent water use data acquisition and control method according to claim 1, characterized in that: The specific process of optimizing feature reliability is as follows: The data of distributed fiber acoustic sensor (DAS) and nano-thin film vibration sensor array are decomposed by wavelet packets to generate three layers of time-frequency sub-bands. The maximum values ​​of the wavelet coefficient modulus are calculated in each of the three layers of time-frequency sub-bands to independently detect the mutation points. The cross-sub-band correlation analysis of the mutation points is used to eliminate the false features of a single frequency band and optimize the feature reliability.

3. The intelligent water use data acquisition and control method according to claim 1, characterized in that: The specific process of extracting effective feature fragments is as follows: Dynamic time warping is used to stretch or compress the time axis of the feature point sequence of the distributed fiber acoustic wave sensor and nanofilm vibration sensor array. The Sakoe-Chiba bandwidth constraint is used to minimize the matching distance, limit the DTW search range, avoid global search, and select turning points on the DTW optimal path as time synchronization markers. The cross power spectral density (CPSD) of the 200ms data segment before and after the synchronization mark point is calculated, and the calculated cross spectral entropy (CSE) value is analyzed. If the cross spectral entropy (CSE) value is greater than 0.85, it is determined to be a valid synchronization feature segment and the valid feature segment is extracted.

4. The intelligent water use data acquisition and control method according to claim 1, characterized in that: The specific process of calculating the temperature influence factor is: The long short-term memory-causal network (LSTM-CN) is used to eliminate temperature interference, perform sliding window processing on the data sequence collected by the distributed optical fiber temperature sensor (DTS), calculate the temperature gradient, and calculate multiple temperature gradients to obtain a temperature gradient sequence. The vibration signal data collected by the distributed optical fiber acoustic sensor (DAS) is subjected to wavelet denoising to extract the vibration energy characteristic sequence. In the input layer of the long short-term memory network LSTM, the temperature gradient sequence and vibration energy feature sequence are input. The temporal relationship between temperature and vibration is captured through the bidirectional LSTM unit in the hidden layer, and the output layer outputs the temperature influencing factor.

5. The intelligent water use data acquisition and control method according to claim 1, characterized in that: The specific method for eliminating the noise introduced by ambient temperature fluctuation is: The formula used is: H′=H·(1-α t ·|T g |) Eliminate the interference of ambient temperature fluctuation on the vibration signal, where α t is the temperature influence factor, which is obtained through dynamic learning of the LSTM network, H is the original vibration energy, T g The temperature gradient represents the change in temperature and is taken in absolute value to focus on the magnitude of the temperature change rather than the direction.

6. The intelligent water use data acquisition and control method according to claim 1, characterized in that: The optimization algorithm of the physical equation constraint is specifically as follows: The PSO algorithm based on the Navier-Stokes equation constraints introduces the pipe material parameters when calculating the pressure wave velocity, calculates the theoretical wave velocity through the formula, and uses the correction coefficient to constrain the particles so that the particle update conforms to the laws of fluid mechanics.

7. The intelligent water use data acquisition and control method according to claim 1, characterized in that: The spatiotemporal domain joint feature matrix data includes: TDOA sequence, vibration energy compensation value, GIS topology-encoded node risk weight, and sensor location coordinates.

8. The intelligent water use data acquisition and control method according to claim 1, characterized in that: The specific process of realizing leak location is as follows: Multimodal feature input and leakage point coordinate inversion, input the spatiotemporal domain joint feature matrix data processed by LSTM-CN and Transformer, and use PSO and PI-GAN for collaborative positioning. PSO updates the particle position according to Navier-Stokes constraints, and PI-GAN filters candidate solutions that violate the wave equation in real time. When the positioning error of the particle swarm optimal solution is less than the threshold, the leakage point coordinates are output.

9. An intelligent water use data acquisition and control system, characterized by: include: Multi-physics field data acquisition module: Arrange sensors to collect multi-physics field data, including: arrival time of the first acoustic wave at the leak point, temperature gradient along the pipeline, vibration signal, and pressure data; Multimodal Data Spatiotemporal Alignment Module: Based on the collected multi-physics field data, the module performs time-frequency decomposition of the acoustic first wave arrival time series and vibration signals including the leak point through multi-scale wavelet transform, detects mutation points to optimize feature reliability, and uses dynamic time warping and cross-spectral entropy to align the time axes of different sensors and extract effective feature segments. Temperature Interference Elimination and Feature Fusion Module: This module uses a long-short-term memory-causal network to analyze the relationship between temperature gradient data and vibration energy in a portion of multi-physics field data after time axis alignment and effective feature extraction. This module then constructs a physical relationship model, calculates the temperature influence factor, and uses this temperature influence factor to compensate for vibration energy, eliminating noise introduced by ambient temperature fluctuations. Physical constraint intelligent positioning module: The vibration energy characteristics, acoustic wave characteristics and time-space joint feature matrix after eliminating temperature interference are used to locate leaks through an optimization algorithm constrained by physical equations and a generative adversarial network.

10. An intelligent water use data acquisition and control terminal device, characterized by: include: Distributed fiber optic acoustic wave sensors, distributed fiber optic temperature sensors, nano-thin film vibration sensor arrays, high-precision pressure sensors, and computer processing units; The computer processing unit is used to execute the computer program of the intelligent water use data acquisition and control method according to any one of claims 1 to 8.

Images