Gas pipe network terminal safety detection system based on internet of things integration

Abstract

The application discloses a gas pipe network end safety detection system based on an internet of things, and belongs to the technical field of gas pipe network safety detection. The system collects pipe wall vibration signals and pipe inner gas pressure fluctuation signals through a piezoelectric sensing array arranged on the outer wall of a pipe section at the end of the gas pipe network, extracts vibration main frequency offset, vibration energy attenuation rate, pressure fluctuation phase difference and pressure pulsation amplitude ratio; maps a state characteristic matrix to a pre-constructed gas pipe network end physical behavior characteristic space, calculates a manifold distance between the state characteristic matrix and each pipe section standard behavior mode characteristic matrix in the space, determines the safety state category of each detection unit according to the manifold distance, and outputs a safety detection result. The system can fully capture the dynamic response difference of multiple physical fields caused by local abnormalities of the pipe section, and realizes high adaptability safety determination in a characteristic space fusing pipe section physical properties.

Description

Technical Field

[0001] This invention relates to the field of gas pipeline safety detection technology, specifically a gas pipeline end safety detection system based on the Internet of Things. Background Technology

[0002] Gas pipelines connect directly to users at their ends, with numerous pipe sections and complex topologies, highlighting the need for real-time detection of abnormal conditions such as leaks and blockages. Existing safety detection methods for gas pipeline ends often employ the method of individually deploying pressure sensors or flow meters at key nodes, inferring the pipeline status by monitoring single macroscopic parameters such as pressure drop and flow rate changes. Some technical solutions introduce vibration sensors, but these are limited to threshold comparisons of vibration amplitude, lacking in-depth analysis of the frequency characteristics and energy attenuation process of vibration signals, making it difficult to distinguish the differences in pipe wall vibration caused by normal operating fluctuations, environmental interference, and genuine minor leaks. The aforementioned detection methods based on single physical quantities or simple statistical characteristics cannot simultaneously utilize the coupled information between the dynamic pressure within the pipe and the dynamic vibration of the pipe wall structure. When a pipe section experiences early leakage or localized blockage due to deposits on the inner wall, both the pressure pulsation characteristics from the gas and the resulting pipe wall vibration characteristics undergo subtle but interconnected changes. These changes are reflected in multiple dimensions, including the drift of vibration frequency over time, the change in the rate of attenuation of vibration energy along the pipe section, and the shift in the phase and amplitude ratio of pressure fluctuations. Because existing methods fail to extract features reflecting the state differences between measuring points at both ends of a pipe segment from these dimensions, they lack sensitivity to early small-scale anomalies, leading to frequent false alarms. Furthermore, the physical properties of pipe segments in the terminal network vary significantly, including length, inner diameter, wall thickness, and number of bends. The same abnormal operating conditions will exhibit different vibration and pressure response patterns on pipe segments with different physical properties. Existing detection methods often apply a uniform feature threshold or metric space to all pipe segments when determining safety status, failing to incorporate the segment's own physical properties into the feature mapping process for state determination. This results in a disconnect between the detection standard and the actual physical behavior of the pipe segments in the terminal areas of the network, where physical properties differ greatly, leading to increased false alarm rates and reduced location accuracy. Summary of the Invention

[0003] This invention provides a gas pipeline network end safety detection system based on the Internet of Things, which is used to solve how to extract multi-dimensional dynamic features reflecting the state differences between the measuring points at both ends of the pipe segment from pipe wall vibration and gas pressure fluctuations in the pipe to sensitively capture early anomalies, and how to integrate the physical properties of each pipe segment into the feature mapping space to achieve accurate safety status determination on pipe segments with different structural parameters.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a gas pipeline network end safety detection system based on the Internet of Things. The system adopts a multi-sensor fusion Internet of Things sensing method to perform fine-grained safety status detection and judgment on each pipe section at the end of the gas pipeline network.

[0005] The system includes a signal acquisition module, a feature extraction module, a vector construction module, a matrix generation module, and a safety determination module. The signal acquisition module uses a piezoelectric sensor array deployed on the outer wall of the end section of the gas pipeline to collect pipe wall vibration signals and gas pressure fluctuation signals at each sensing node. Along the gas delivery direction, the piezoelectric sensor array integrates piezoelectric accelerometers and piezoresistive pressure sensors on the outer wall surfaces at both ends of each pipe section, simultaneously acquiring pipe wall vibration acceleration time-history data and gas pressure time-history data with a unified sampling frequency and duration. The vibration acceleration time-history data is integrated twice to obtain the pipe wall vibration displacement signal, and the pressure time-history data is digitally low-pass filtered to obtain the gas pressure fluctuation signal.

[0006] The feature extraction module extracts the vibration dominant frequency offset and vibration energy attenuation rate from the pipe wall vibration signal at each sensing node, and extracts the pressure fluctuation phase difference and pressure pulsation amplitude ratio from the gas pressure fluctuation signal inside the pipe. Preferably, a short-time Fourier transform is performed on the pipe wall vibration displacement signal, and the frequency component with the largest amplitude in the vibration spectrum at each moment is taken as the instantaneous vibration dominant frequency. The slope of the linear trend of the instantaneous vibration dominant frequency time series is calculated to obtain the vibration dominant frequency offset. The time-domain energy envelope is extracted from the pipe wall vibration displacement signal, and the exponential decay coefficient of the vibration signal amplitude envelope is obtained as the vibration energy attenuation rate. The gas pressure fluctuation signal inside the pipe at the same sensing node is bandpass filtered to extract the pressure fluctuation component within a preset frequency band. The instantaneous phase sequence and instantaneous amplitude sequence of the pressure fluctuation component are obtained through Hilbert transform. The slope of the instantaneous phase sequence linearly fitted along the time axis is taken as the pressure fluctuation phase difference, and the ratio of the time average value of the instantaneous amplitude sequence to the static pressure of the gas inside the pipe is taken as the pressure pulsation amplitude ratio.

[0007] The vector construction module uses the pipe segment between two adjacent sensing nodes as the detection unit. It constructs a multivariate state feature vector for this detection unit using the differences in vibration dominant frequency offset, vibration energy attenuation rate, pressure fluctuation phase difference, and pressure pulsation amplitude ratio between the sensing nodes at both ends of the detection unit. Specifically, the module subtracts the vibration dominant frequency offset of the upstream sensing node from the vibration dominant frequency offset of the downstream sensing node to obtain the dominant frequency offset difference; subtracts the vibration energy attenuation rates of the upstream and downstream nodes to obtain the energy attenuation rate difference; subtracts the pressure fluctuation phase difference of the upstream and downstream nodes to obtain the phase difference difference; and subtracts the pressure pulsation amplitude ratio of the upstream and downstream nodes to obtain the amplitude ratio difference. These differences are then combined in a fixed-dimensional order to form the multivariate state feature vector for the corresponding detection unit.

[0008] The matrix generation module arranges the multivariate state feature vectors of all detection units according to the pipe segment topology order to form a state feature matrix at the end of the gas pipeline network. As a technical solution of this invention, the order of all detection units is determined according to the actual gas flow direction. The multivariate state feature vector of each detection unit is used as a row of the matrix. An initial state feature matrix is ​​constructed using the arrangement order as the row index and the vector dimension components as the column index. Each element in the initial state feature matrix undergoes deviation standardization to map the numerical range to a predetermined interval. The processed matrix is ​​then used as the state feature matrix.

[0009] The safety determination module maps the state feature matrix to a pre-constructed physical behavior feature space at the end of the gas pipeline network. Within this physical behavior feature space, it calculates the manifold distance between the state feature matrix and the standard behavior pattern feature matrix of each pipe segment. Based on the manifold distance, it determines the safety status category of each detection unit and outputs the safety detection result. During the mapping process, the physical attribute parameters of each pipe segment at the end of the gas pipeline network are obtained, including segment length, inner diameter, wall thickness, and number of bends. These physical attribute parameters are used to construct the physical attribute vector corresponding to each pipe segment. Using the physical attribute vectors of all pipe segments as basis vectors, a linear transformation projection is performed on the state feature matrix, mapping it from the original feature dimension space to the physical behavior feature space spanned by the basis vectors, thus obtaining the projected coordinate matrix. During manifold distance calculation, standard state feature matrices for each pipe segment under normal, leakage, and blockage conditions are extracted from the historical database. These standard state feature matrices are then mapped to the aforementioned physical behavior feature space to obtain the standard projected coordinate matrix for each pipe segment under each condition. In the physical behavior feature space, the geodesic distance between the projected coordinate matrix of the state feature matrix and the standard projected coordinate matrix for each current pipe segment under each condition is calculated and used as the manifold distance. For safety status category determination, all conditions corresponding to the current pipe segment are traversed, and the manifold distance values ​​between the projected coordinate matrix and the standard projected coordinate matrix for each condition are obtained. All manifold distance values ​​are sorted in ascending order, and the condition corresponding to the minimum manifold distance value is marked as the preliminary determination condition. The minimum manifold distance value is compared with a preset distance threshold. If the minimum manifold distance value is less than or equal to the distance threshold, the preliminary determination condition is output as the safety status category for that detection unit.

[0010] Furthermore, the safety determination module sequentially obtains the safety status category of each detection unit along the pipeline topology, generates a safety status category sequence, continuously identifies all detection units marked as leaking conditions in the sequence, merges the continuously arranged leaking condition detection units with a number greater than a preset threshold into a leaking area, and finally summarizes and outputs the location information of the leaking area, the number of detection units it contains, and the safety status category of each detection unit as the safety detection result of the gas pipeline network end.

[0011] The aforementioned system utilizes a piezoelectric sensor array to achieve synchronous IoT acquisition of pipe wall vibration and pressure fluctuations. Through multi-feature extraction and vector construction, it comprehensively represents the differentiated state changes of each detection unit at the end of the pipeline network into a state feature matrix. By introducing the physical attributes of the pipe segments to construct a physical behavior feature space, and achieving accurate matching with standard behavior patterns based on manifold distance metrics within this space, it can sensitively identify abnormal operating conditions such as leaks and blockages. Furthermore, it can perform regional merging and localization of continuous leaks in multiple pipe segments within the pipeline network, thereby improving the accuracy and interpretability of safety detection at the end of the gas pipeline network.

[0012] The technical effects and advantages provided by the present invention in the above technical solution are as follows: A multivariate state feature vector for the detection unit is constructed using the differences in vibration dominant frequency offset, vibration energy attenuation rate, pressure fluctuation phase difference, and pressure pulsation amplitude ratio between adjacent sensing nodes. The vibration dominant frequency offset characterizes the difference in the time-varying trend of vibration frequencies at both ends of the pipe section, reflecting the asymmetric performance in the frequency dimension caused by changes in local structural stiffness or flow-induced vibration coupling conditions due to leakage or blockage. The vibration energy attenuation rate difference reflects the difference in the rate of attenuation of vibration energy at both ends during propagation along the pipe section; when micro-cracks appear in the pipe wall or local changes occur in the impedance of the medium inside the pipe, the energy dissipation path changes, resulting in inconsistent attenuation rates upstream and downstream. The pressure fluctuation phase difference extracts the change in time delay of pressure wave propagation at both ends of the pipe section, and is sensitive to wave velocity abrupt changes caused by local blockage or wave field distortion caused by leakage. The pressure pulsation amplitude ratio difference reflects the relative change in pulsating energy at both ends, and can capture the redistribution of pressure pulsating energy within the pipe section caused by abnormal operating conditions. These four differences are combined into a multivariate state feature vector. Simultaneously, the asymmetry of the dynamic response at both ends of the detection unit is comprehensively characterized from four dimensions: frequency offset, energy attenuation, phase delay, and amplitude ratio. This ensures that subtle signal changes are not overwhelmed by overlapping fluctuations of a single feature, enabling sensitive identification of weak two-port differences caused by leaks or blockages at an early stage. The state feature matrix is ​​mapped to a physical behavior feature space spanned by pipe segment length, inner diameter, wall thickness, and the number of bends. Within this space, the manifold distance between the state feature matrix and the standard behavior mode feature matrix of each pipe segment is calculated to determine the safety state category. Pipe segment length and inner diameter determine the flow inertia and pressure wave propagation time of the gas; wall thickness affects the vibration stiffness of the pipe wall and its energy storage capacity against external excitation; and the number of bends is directly related to the local resistance of the flow channel and the complexity of the flow-induced vibration modes. The basis vectors, composed of these physical attribute vectors, can express the inherent physical behavior constraints of the pipe segment. After projecting the original state feature matrix onto the physical behavior feature space spanned by these basis vectors, the features measured on pipe segments with different physical attributes are placed under a unified behavioral reference framework, eliminating the differences in feature response benchmarks caused by different pipe segment structural parameters. The geodesic distance between the projected coordinate matrices in the physical behavior feature space is calculated as the manifold distance. The geodesic measurement follows the inherent geometric structure of the feature data in this space and has adaptive fitting capabilities for the nonlinear shape of the feature distribution. This makes the distance measurement between the standard behavioral pattern feature matrix and the current state feature matrix under normal operating conditions, leakage conditions, and blockage conditions more closely reflect the actual physical processes at the end of the pipeline network. The manifold distance-based determination method can maintain the stability and accuracy of safety status category identification at the end of gas pipeline networks with large differences in pipe segment physical attributes, reducing false alarms caused by inherent structural differences in pipe segments. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0014] Figure 1 This is a schematic diagram of a gas pipeline network end-of-line safety detection system based on the Internet of Things. Figure 2 This is a flowchart of synchronous acquisition and signal processing of vibration and pressure at the end of the gas pipeline network; Figure 3 This is a flowchart of the zero-phase filtering and Hilbert transform processing of the gas pressure fluctuation signal inside the pipe. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] See Figure 1This invention provides a gas pipeline network end-point safety detection method based on the Internet of Things. The system implementing this method includes a signal acquisition module, a feature extraction module, a vector construction module, a matrix generation module, and a safety determination module. The signal acquisition module collects pipe wall vibration signals and gas pressure fluctuation signals at each sensing node using a piezoelectric sensor array deployed on the outer wall of the end section of the gas pipeline network. The feature extraction module extracts the vibration dominant frequency offset and vibration energy attenuation rate from the pipe wall vibration signals at each sensing node, and extracts the pressure fluctuation phase difference and pressure pulsation amplitude ratio from the gas pressure fluctuation signals. The vector construction module uses the pipe section between two adjacent sensing nodes as a detection unit, and constructs a multivariate state feature vector for the detection unit using the difference in vibration dominant frequency offset, vibration energy attenuation rate, pressure fluctuation phase difference, and pressure pulsation amplitude ratio between the sensing nodes at both ends of the detection unit. The matrix generation module arranges the multivariate state feature vectors of all detection units according to the pipe section topology to form a state feature matrix at the end of the gas pipeline network. The safety determination module maps the state feature matrix to a pre-constructed physical behavior feature space at the end of the gas pipeline network, calculates the manifold distance between the state feature matrix and the standard behavior pattern feature matrix of each pipe section in the physical behavior feature space, determines the safety status category of each detection unit based on the manifold distance, and outputs the safety detection result.

[0017] Example 1: In specific implementation, please refer to Figure 2 Along the gas delivery direction at the end of the gas pipeline network, the start and end positions of each pipe segment are determined. A piezoelectric accelerometer and a piezoresistive pressure sensor are installed on the outer wall of each pipe segment at both the starting and ending points. The piezoelectric accelerometer is installed by using a high-temperature resistant epoxy resin adhesive to attach the sensor base to the outer surface of the pipe wall, applying a constant preload to maintain mechanical coupling rigidity, ensuring the sensor's sensitive axis is perpendicular to the pipe wall's tangential direction. The piezoresistive pressure sensor is installed by pre-drilling pressure tapping holes at the start and end points of the pipe segment and welding internally threaded pressure tapping connectors. The threaded end of the piezoresistive pressure sensor is then screwed into the pressure tapping connector and sealed tightly, allowing the pressure-sensing diaphragm of the piezoresistive pressure sensor to directly contact the gas inside the pipe.

[0018] All piezoelectric accelerometers and piezoresistive pressure sensors installed at the beginning and end of each pipe section are arranged into a piezoelectric sensor array. All sensors are connected to a centralized data acquisition unit via shielded cables of the same type. The data acquisition unit uses a temperature-controlled crystal oscillator inside to provide a clock reference and sends the same sampling trigger signal to all channels to achieve synchronous sampling. A uniform sampling frequency of 2000 Hz and a uniform sampling duration of 600 seconds are set. The sampling frequency was selected based on the following criteria: According to NB / T 10620-2021 "Vibration Testing of Pressure Equipment", the sampling frequency should not be less than 2.56 times the upper limit of the main frequency range; referring to GB 50868-2013 "Permissible Vibration Standard for Building Engineering", the ratio of sampling frequency to cutoff frequency should be 2.5 to 6.0; and considering the main frequency range (0 to 500 Hz) of the pipe wall vibration at the end of the gas pipeline network and the high-frequency vibration components that may be caused by abnormal conditions such as leakage and blockage, the sampling frequency was determined to be 2000 Hz to satisfy the Nyquist sampling theorem and fully capture the frequency domain characteristics of the pipe wall vibration signal.

[0019] The sampling duration was selected based on the following criteria: a sampling duration of 600 seconds provides approximately 1171 short-time Fourier transform time windows (1024-point Hanning window, 256-step), meeting the statistical stability requirements of spectral analysis; it also covers a sufficient number of pressure fluctuation cycles (dominant frequency 0.5–5 Hz) to extract stable time-frequency domain features; referring to the engineering practice of pipeline monitoring based on vibration signals, the sample length is (1–1.5) × 10 5 The sampling points are reasonable, with 600 seconds corresponding to 1.2 × 10⁻⁶. 6 The sampling points are sufficient to meet the data requirements for subsequent manifold distance calculation and pattern recognition. During the sampling process, piezoelectric accelerometers simultaneously collect time-history data of pipe wall vibration acceleration at each sensing node, and piezoresistive pressure sensors simultaneously collect time-history data of gas pressure inside the pipe at each sensing node.

[0020] For each channel of pipe wall vibration acceleration time history data, a quadratic integration process is performed to obtain the pipe wall vibration displacement signal. The quadratic integration process first uses the least squares method to remove the linear trend term from the pipe wall vibration acceleration time history data. The acceleration time history data after removing the trend term is denoted as... ,in The sampling point number, , The total number of sampling points. The velocity time history is calculated using the trapezoidal integral method in the time domain. :

[0021] in, The sampling time interval, Second; For the first Detrended acceleration values ​​at each sampling point; For the first Detrended acceleration values ​​at each sampling point; For the first Velocity values ​​at each sampling point, initial velocity Set to 0 m / s. For the velocity time history... After removing the linear trend term again, the displacement time history was calculated using the same trapezoidal integration method. Displacement time history This serves as the pipe wall vibration displacement signal at the corresponding sensing node.

[0022] For each pipeline gas pressure time-history data point, a digital low-pass filter is applied to obtain the pipeline gas pressure fluctuation signal. The digital low-pass filter is an infinite impulse response digital filter, specifically a fourth-order Butterworth low-pass filter, with a cutoff frequency set to 15 Hz. The selection of the cutoff frequency is based on the following criteria: First, the pressure fluctuation component within the 0.5–5 Hz frequency band needs to be extracted in subsequent signal processing. The cutoff frequency of the low-pass filter should be higher than the upper limit of the effective frequency of the target signal, typically with a design margin of 2–5 times. 15 Hz is approximately 3 times that of 5 Hz, ensuring that the signal amplitude within the 0.5–5 Hz target frequency band is not attenuated. Second, the energy of the pressure fluctuation signal at the end of the gas pipeline network is mainly concentrated in the low-frequency band of 0–30 Hz. A cutoff frequency of 15 Hz can completely preserve the main frequency energy of the pressure fluctuation signal. Third, this cutoff frequency is much lower than half of the sampling frequency (2000 Hz) (i.e., 1000 Hz), effectively preventing high-frequency noise from aliasing into the low-frequency band. Finally, a cutoff frequency of 15 Hz can effectively filter out non-leakage-related high-frequency interference components such as sensor electronic noise, environmental electromagnetic interference, and high-frequency pulsation of pipeline turbulence. The time history data of the gas pressure in the pipe is input into a digital low-pass filter. The output of the differential equation of the filter is the pressure fluctuation component after removing high-frequency noise. This pressure fluctuation component is used as the gas pressure fluctuation signal in the pipe at the corresponding sensing node.

[0023] Example 2: In practice, a short-time Fourier transform is performed on the pipe wall vibration displacement signal acquired at each sensing node. The short-time Fourier transform process is as follows: a Hanning window function with a length of 1024 sampling points is selected as the window function, and the step size of each sliding step of the window function is set to 256 sampling points. The selection criteria for window length and step size are as follows: First, at a sampling frequency of 2000Hz, a window length of 1024 points corresponds to a time window of 0.512 seconds, which provides sufficient frequency resolution to distinguish subtle shifts in the dominant vibration frequency without excessively sacrificing time resolution. Furthermore, 1024 is an integer power of 2, which fully leverages the computational efficiency advantage of the Fast Fourier Transform (FFT). This window length is also one of the commonly used standards in engineering practice for vibration signal spectrum analysis. Second, the Hanning window is one of the standard time-domain window functions specified in "Mechanical Vibration and Shock Signal Processing Part 2: Time-Domain Window for Fourier Transform Analysis," which has the advantages of low sidelobe levels and low spectral leakage, making it suitable for preprocessing of vibration signal spectrum analysis. Third, a step size of 256 sampling points is approximately 1 / 4 of the window length (overlap rate of 75%), which is one of the most commonly used overlap settings in short-time Fourier transform. This effectively controls the computational load while ensuring time resolution to meet the subsequent requirements for calculating the linear trend slope of the dominant vibration frequency shift, and ensures a smooth transition of information between adjacent windows, avoiding the loss of signal features at window boundaries. The pipe wall vibration displacement signal is sequentially segmented into multiple displacement signal segments according to the sliding step size. A discrete Fourier transform is calculated for each displacement signal segment after windowing to obtain the corresponding spectrum vector. The spectrum vectors of the displacement signal segments at all times for the same sensing node are arranged in time sequence to form a vibration spectrum time history diagram. In the vibration spectrum at each time moment, the frequency component with the largest amplitude is searched, and the frequency point value of this frequency component is marked as the instantaneous vibration dominant frequency at the corresponding time moment.

[0024] Arrange the instantaneous dominant frequencies of vibration at all moments in chronological order to form a one-dimensional time series. Use the moment indices within this time series as independent variables and the instantaneous dominant frequency value at each moment as the dependent variable, then perform least-squares linear fitting. Least-squares linear fitting requires solving for the coefficients of the linear model, whose expression is:

[0025] in, For the first The time value corresponding to each moment. The range of values ​​is , The total length of the time series of the instantaneous dominant vibration frequency; In time The instantaneous vibration dominant frequency fitting value obtained by linear fitting; The slope of the linear fit is used as the slope of the desired linear trend. This represents the intercept of the linear fit. During the solution process, the slope is calculated using the normal equation method. The estimated value will be used to calculate the slope. The estimated value is used as the vibration dominant frequency offset at the corresponding sensing node.

[0026] In practical implementation, time-domain energy envelope extraction is performed on the pipe wall vibration displacement signal at the same sensing node. The processing procedure for time-domain energy envelope extraction is as follows: First, a Hilbert transform is performed on the pipe wall vibration displacement signal to obtain the analytic signal. Then, the modulus value of the analytic signal is taken to obtain a discrete sequence of the displacement signal amplitude envelope. The technical principle and significance of this step are as follows: The pipe wall vibration displacement signal contains high-frequency oscillation components, making it difficult to observe the macroscopic trend of energy decay through direct analysis. The Hilbert transform can convert a one-dimensional real signal into a two-dimensional analytic signal, and the modulus value of the analytic signal is the instantaneous amplitude envelope of the original signal. This envelope removes the high-frequency oscillation components from the vibration signal while preserving the decay profile of the vibration energy over time, making it a standard method for vibration signal envelope demodulation analysis. By extracting the envelope, the energy decay trend of the original vibration signal can be separated from the high-frequency oscillation components, providing a reliable amplitude decay data basis for the accurate calculation of the subsequent exponential decay coefficient. If the original vibration signal is directly subjected to exponential fitting, the high-frequency oscillation components will severely interfere with the fitting accuracy, making it impossible to obtain a stable estimate of the decay coefficient. In the fields of pipeline vibration analysis and mechanical fault diagnosis, Hilbert transform envelope extraction has been widely applied to demodulation analysis of fault-modulated vibration signals and vibration feature extraction of pipeline leakage signals. A logarithmic transformation is performed on the discrete sequence of the displacement signal amplitude envelope to convert the exponential decay pattern into a linear decay pattern. In the logarithmic domain, least-squares linear fitting is used to fit the data points of the descending segment of the envelope in the logarithmic domain, obtaining the slope value of the fitted line. This slope value is the estimated value of the exponential decay coefficient. The calculated estimated value of the exponential decay coefficient is used as the vibration energy attenuation rate at the corresponding sensing node. The descending segment data points of the envelope are defined as follows: First, all data points in the discrete sequence of the amplitude envelope are traversed. Starting from the beginning of the sequence, the first data point with an amplitude lower than the global maximum amplitude of the envelope is determined as the starting point of the descending segment. Then, the search continues from this starting point. When the amplitude of several consecutive adjacent data points no longer shows a decreasing trend, and the amplitude rebounds or stabilizes without significant decrease, the last data point in a decreasing amplitude state before the rebound or stabilization is taken as the end point of the descending segment. All data points from the starting point to the ending point are the data points of the descending segment of the envelope. Only the data points within this interval are selected for subsequent least squares linear fitting calculations in the logarithmic domain. Data from the oscillating and stationary baseline portions outside the descending segment are removed to avoid interference from non-decaying interval data in solving the exponential decay coefficient.

[0027] Example 3: In specific implementation, please refer to Figure 3 For the gas pressure fluctuation signal collected at each sensing node, bandpass filtering is performed to extract the pressure fluctuation component within a preset frequency band. The bandpass filtering uses a zero-phase digital bandpass filter, which is composed of two cascaded Butterworth filters of the same order. Forward and backward filtering are performed once each to eliminate phase distortion. The lower cutoff frequency of the preset frequency band is set to 0.5 Hz, the upper cutoff frequency to 5 Hz, the passband ripple factor to 0.1 dB, and the stopband attenuation to 40 dB. The selection criteria for these filter parameters are as follows: Frequency band (0.5 Hz to 5 Hz): The energy of pressure fluctuation signals at the end of the gas pipeline network is mainly concentrated in the low-frequency band of 0–30 Hz, where the dominant frequency of periodic pressure pulsations caused by pressure regulating equipment operation and user load changes is usually below 10 Hz. The lower cutoff frequency of 0.5 Hz is used to filter out pressure baseline drift (usually below 0.5 Hz) caused by factors such as slow changes in ambient temperature and daily load cycle changes; the upper cutoff frequency of 5 Hz is used to filter out high-frequency interference components such as pipeline turbulence noise and sensor electronic noise (usually above 5 Hz). This frequency band can effectively cover the dominant frequency components of pressure fluctuation modulation information related to abnormal operating conditions such as leakage and blockage. In addition, when using Hilbert transform for envelope demodulation analysis, the passband range of the bandpass filter must include the main frequency components of the modulated signal. The bandwidth of 0.5 Hz to 5 Hz is approximately 3.32 octaves, which can provide a signal basis with sufficient periodic stability for subsequent linear fitting of instantaneous phase sequences, ensuring the calculation accuracy of pressure fluctuation phase difference and pressure pulsation amplitude ratio. Passband ripple (0.1 dB): Passband ripple refers to the maximum deviation of the filter's amplitude-frequency response within the passband from the ideal flat response. Stopband attenuation (40 dB): Stopband attenuation refers to the filter's ability to suppress signals at frequencies outside the passband. A stopband attenuation of 40 dB means that signals outside the passband are attenuated to 1% of their original value, effectively suppressing baseline drift below 0.5 Hz and high-frequency interference above 5 Hz, providing a sufficiently clean input signal for the Hilbert transform. The time series of the gas pressure fluctuation signal in the pipe is input into a zero-phase digital bandpass filter. After two stages of forward filtering and backward filtering, the filtered pressure fluctuation component sequence is output. This pressure fluctuation component sequence is used as the pressure fluctuation component extracted at the corresponding sensing node.

[0028] The extracted pressure fluctuation components are subjected to a Hilbert transform to obtain their instantaneous phase and amplitude sequences. The Hilbert transform process is as follows: First, a Fast Fourier Transform (FFT) is performed on the discrete-time sequence of the pressure fluctuation components to obtain their spectral sequence. The spectral values ​​of the negative frequency components are set to zero, and the spectral values ​​of the positive frequency components are multiplied by a coefficient of 2 to obtain the Hilbert-transformed spectral sequence. An Inverse Fast Fourier Transform (IFFT) is then performed on the Hilbert-transformed spectral sequence to obtain the analytic signal sequence of the pressure fluctuation components. In the analytic signal sequence, the real part of each complex number represents the original pressure fluctuation component value, and the imaginary part represents the Hilbert transform result. The principal argument and magnitude of each complex number in the analytic signal sequence are calculated. The principal argument values ​​of all complex numbers are arranged in chronological order to form the instantaneous phase sequence, and the magnitudes of all complex numbers are arranged in chronological order to form the instantaneous amplitude sequence.

[0029] Linear fitting is performed along the time axis on the instantaneous phase sequence to obtain the pressure fluctuation phase difference. The linear fitting process is as follows: the time index corresponding to each sampling point in the instantaneous phase sequence is used as the independent variable, and the instantaneous phase value of each sampling point in the instantaneous phase sequence is used as the dependent variable. The slope and intercept of the fitted line are calculated using the least squares method. The least squares method is solved using the normal equation, and the least squares estimate of the slope is obtained through matrix operations. The slope of the calculated fitted line is used as the pressure fluctuation phase difference at the corresponding sensing node.

[0030] In practical implementation, the time average of the instantaneous amplitude sequence is calculated by summing the instantaneous amplitudes of all sampling points in the sequence and dividing by the total number of sampling points. The static pressure of the gas inside the pipe at the corresponding sensing node is then obtained. This static pressure is acquired by taking the pressure data from the initial 10 seconds of the time-history data collected by the piezoresistive pressure sensor and calculating its arithmetic mean. This arithmetic mean is then used as the static pressure of the gas inside the pipe. The time average is then divided by the static pressure of the gas inside the pipe, and the resulting ratio is used as the pressure pulsation amplitude ratio at the corresponding sensing node.

[0031] Example 4: In practical implementation, for each pipe segment at the end of the gas pipeline network, the sensor nodes installed at both ends are identified as the upstream and downstream sensor nodes of that segment, respectively. The identification is based on the gas delivery direction at the end of the gas pipeline network: the sensor node through which the gas flows first is determined as the upstream sensor node, and the sensor node through which the gas flows last is determined as the downstream sensor node. Taking the pipe segment between two adjacent sensor nodes as a detection unit, for each detection unit, the vibration dominant frequency offset of its upstream and downstream sensor nodes is obtained. The vibration dominant frequency offset of the upstream and downstream sensor nodes is subtracted, and the result is taken as the dominant frequency offset difference of the detection unit. The vibration energy attenuation rate of the upstream and downstream sensor nodes is also obtained and subtracted, and the result is taken as the energy attenuation rate difference of the detection unit. The pressure fluctuation phase difference between the upstream and downstream sensing nodes is obtained. The phase difference between the upstream and downstream sensing nodes is subtracted, and the result is used as the phase difference value of the detection unit. The pressure pulsation amplitude ratio between the upstream and downstream sensing nodes is also obtained. The amplitude ratio difference between the upstream and downstream sensing nodes is subtracted, and the result is used as the amplitude ratio difference value of the detection unit.

[0032] The differences in main frequency offset, energy attenuation rate, phase difference, and amplitude ratio are combined in a preset fixed-dimensional order to form a multi-dimensional state feature vector for the detection unit. The fixed-dimensional order is defined as follows: the first dimension of the multi-dimensional state feature vector is the main frequency offset difference, the second dimension is the energy attenuation rate difference, the third dimension is the phase difference, and the fourth dimension is the amplitude ratio difference. This operation is performed on each detection unit to obtain a four-dimensional multi-dimensional state feature vector for each unit.

[0033] In practice, the order of all detection units is determined according to the actual flow direction of the gas within the end of the gas pipeline network. Starting from the initial pipe segment where the gas enters the end of the gas pipeline network, each pipe segment is sequentially traversed along the gas flow path, and the detection units corresponding to each segment are numbered sequentially to obtain the sequence of detection units. The order of the detection units in the sequence is used as the row index, and the row index is... , The range of values ​​is ,in, This represents the total number of detection units contained at the end of the gas pipeline network. The column indices are based on the dimensional components of the multivariate state feature vector. , The value is The corresponding dimension of the main frequency offset difference. The value is The corresponding dimension is the difference in energy decay rate. The value is The corresponding phase difference dimension The value is The corresponding amplitude ratio difference dimension. The first... The first multivariate state feature vector of the detection unit Dimension component values, filled into the initial state feature matrix. Line number Column position, traverse all and After that, a OK The initial state feature matrix of the column.

[0034] For each element in the initial state feature matrix, deviation normalization is performed so that the value of each element is mapped to... Interval. The specific method for deviation standardization is as follows: for each column index of the initial state feature matrix. Extract all columns in this column. Given a column of element values, retrieve the minimum value. and maximum value . The method of obtaining is to take the first List all Each element value is compared according to its numerical value, and the smallest value is selected as the smallest value. ; The method of obtaining is to take the first List all Each element value is compared numerically, and the largest value is selected as the [value]. For the first Each element in the column Calculate the standardized deviation value according to the following formula. :

[0035] in, The first element in the standardized deviation matrix Line number The element values ​​of the column; The first element in the initial state feature matrix Line number The original element values ​​of the column; The initial state feature matrix is ​​the first... The minimum value in the column; The initial state feature matrix is ​​the first... The maximum value of a column. When a column's maximum value is... and When the values ​​are equal, the first Standardized deviations of all elements in the column Set the value to For all Lines and all After the above calculations are completed, all columns will be... Composition OK The column matrix is ​​the state feature matrix.

[0036] Example 5: In practical implementation, the physical property parameters of each pipe segment at the end of the gas pipeline network are obtained. These parameters include pipe segment length, inner diameter, wall thickness, and number of bends. The pipe segment length is obtained by extracting the axial length data from the design or as-built drawings of the gas pipeline network. If the drawing data is missing, a handheld laser rangefinder is used to measure the actual pipe segment along its route on-site. The measurement result is rounded to two decimal places in meters. The inner diameter is obtained by reading the nominal inner diameter value from the pipe segment's nameplate or material certificate. If the nameplate or certificate is missing, an ultrasonic thickness gauge is used to measure the wall thickness in four quadrants at three cross-sections at both ends and the middle of the pipe segment. The measured inner diameter is obtained by subtracting the measured outer diameter from the average wall thickness. The average inner diameter of the three cross-sections is taken as the inner diameter of the pipe segment, in millimeters. The pipe section wall thickness is obtained by using an ultrasonic thickness gauge to measure the wall thickness in four quadrants at three cross-sections at both ends and the middle of the pipe section. The average of the twelve wall thickness measurements is taken as the pipe section wall thickness, in millimeters. The number of bends in the pipe section is obtained by conducting an on-site survey along the pipe section axis and counting the number of bends between the start and end points of the pipe section. Bends include 90-degree bends, 45-degree bends, and reducing bends. Each change in pipe direction or pipe diameter is counted once.

[0037] A physical attribute vector is constructed for each pipe segment using its length, inner diameter, wall thickness, and number of bends. For each pipe segment at the end of the gas pipeline network, the segment length is used as the first component of the physical attribute vector, the inner diameter as the second component, the wall thickness as the third component, and the number of bends as the fourth component. These four components are arranged in a fixed order to form a four-dimensional vector, which is the physical attribute vector of the corresponding pipe segment. This process is repeated for all pipe segments at the end of the gas pipeline network, resulting in the same number of physical attribute vectors as the total number of pipe segments.

[0038] Using the physical attribute vectors of all pipe segments as basis vectors, a linear transformation projection is performed on the state feature matrix to map it from the original feature dimension space to the physical behavior feature space spanned by the basis vectors. The linear transformation projection process is as follows: The physical attribute vectors of all pipe segments are arranged row-wise to form a basis matrix. The number of rows in the basis matrix equals the total number of pipe segments, and the number of columns is 4. The basis matrix is ​​then subjected to Schmitt orthogonalization to obtain a set of orthonormal basis vectors. The number of orthonormal basis vectors equals the column rank of the basis matrix. Each row vector of the state feature matrix is ​​projected onto each orthonormal basis vector using an inner product. The inner product result serves as the coordinate value of that row vector in the new feature dimension. After all row vectors are projected, the projected coordinate matrix of the state feature matrix in the physical behavior feature space is obtained.

[0039] In practical implementation, standard state feature matrices for each pipe section under normal, leakage, and blockage conditions are extracted from the historical database. The historical database is constructed as follows: after the gas pipeline network is put into operation at the end, pipe wall vibration signals and gas pressure fluctuation signals within the pipe are continuously collected from each sensor node according to a set collection cycle. Based on the collected signals, multivariate state feature vectors and state feature matrices for each detection unit are calculated. During each collection, the actual operating condition label of the pipe section is recorded synchronously. The operating condition labels are derived from event records of the pipeline monitoring system or manual inspection results, and are divided into three types: normal operating condition label, leakage operating condition label, and blockage operating condition label. The selection criteria for these three operating condition classifications are as follows: According to the "Basic Terminology Standard for Urban Gas Engineering," gas pipeline operating conditions are divided into design operating conditions, operating conditions, and accident operating conditions. Accident operating conditions include situations such as pipeline leakage caused by external force damage. In the field of gas pipeline fault diagnosis and safety inspection, normal, leakage, and blockage conditions are the three most basic and commonly used operating condition classifications. In this technical solution, the normal operating condition serves as a benchmark, providing a baseline for the characteristic behavior patterns of each pipe section. The leakage condition corresponds to pressure wave field distortion and changes in energy dissipation paths caused by pipe wall cracks or damage. The blockage condition corresponds to changes in flow resistance and sudden changes in wave velocity caused by the accumulation of localized deposits or impurities on the pipe wall. Leakage and blockage are considered as the two main abnormal operating conditions for safety status determination because they are based on the two most common safety risk types at the end of gas pipeline networks: leakage leads to gas escape and explosion risk, while blockage leads to gas supply interruption or pressure abnormalities. Furthermore, these two conditions have different characteristic response patterns at the physical level and need to be modeled and identified separately. Therefore, they are used as the target operating condition categories for safety status determination in this system. The state feature matrix obtained from multiple acquisitions of the same pipe section under the same operating condition label is used as the standard state feature matrix group for that pipe section and operating condition. The average value of all matrices in the standard state feature matrix group is taken element-wise to obtain an average standard state feature matrix. This average standard state feature matrix is ​​then stored as the standard state feature matrix corresponding to that pipe section and operating condition.

[0040] Perform the same linear transformation projection operation as the state feature matrix on the standard state feature matrix corresponding to the normal operating condition, the standard state feature matrix corresponding to the leakage condition, and the standard state feature matrix corresponding to the blockage condition on the standard state feature matrix. That is, multiply each standard state feature matrix by the same set of standard orthogonal basis vectors to obtain the standard projected coordinate matrix of each pipe segment under normal operating condition, the standard projected coordinate matrix of the leakage condition, and the standard projected coordinate matrix of the blockage condition.

[0041] In the physical behavior feature space, the geodesic distance between the projected coordinate matrix of the state feature matrix and the standard projected coordinate matrix of each working condition of the current pipe segment is calculated, and this geodesic distance is used as the manifold distance. The calculation of the geodesic distance adopts the shortest path graph approximation method in the equidistant feature mapping algorithm. The specific calculation process is as follows: the first... Each row vector is denoted as , , The total number of rows in the projected coordinate matrix; the first row of the standard projected coordinate matrix. Each row vector is denoted as , , This represents the total number of rows in the standard projected coordinate matrix. Construct a joint dataset containing all row vectors. and all row vectors common There are 10 data points. For each data point in the joint dataset, find the nearest Euclidean distance to it. Adjacent data points, The value is At each data point and its Edges are established between adjacent data points, with the weight of each edge being the Euclidean distance between the two points, to construct a network containing... The adjacency graph structure of the data points. The selection of the K value (neighborhood size) is based on the following: In the Isomap algorithm, the neighborhood size K is the only free parameter that needs to be manually specified, and its value directly affects the connectivity of the neighborhood graph and the estimation accuracy of the geodesic distance. If the K value is too small, the neighborhood graph will break into multiple disconnected components, making it impossible to calculate the shortest path between some data points; if the K value is too large, short edges will be generated, incorrectly connecting data points that are far apart on the manifold, resulting in distorted geodesic distance estimation. In this technical solution, the row vector dimension of the projected coordinate matrix of the state feature matrix is ​​composed of four physical attribute parameters: pipe segment length, inner diameter, wall thickness, and number of bends, that is, the intrinsic dimension of the data manifold d=4, and the K value needs to be greater than d to maintain the stability of the dataset's topological structure. After comprehensively considering factors such as neighborhood graph connectivity, avoiding short-circuit effects, computational efficiency, and data scale, K=12 was selected as an empirical value. This value is in the upper middle of the empirical range of 5-15 typically used in manifold learning engineering applications, ensuring sufficient connectivity of the neighborhood graph while effectively controlling computational load and avoiding the introduction of short-circuit edges. The Dijkstra shortest path search algorithm is run on the adjacency graph structure to calculate the shortest path distance between each pair of data points in the joint dataset. This algorithm uses the weights of each edge in the adjacency graph (i.e., the Euclidean distance between each pair of data points) as the basis for path search. By calculating the sum of the weights of all possible paths between two points, the path with the smallest weight is selected as the shortest path. The sum of the weights of this shortest path is the estimated geodesic distance between the two points. After traversing all data point pairs, the following is obtained: OK The shortest path distance matrix for each column. Extract the row index range from the shortest path distance matrix. to The column index range is to The submatrix of the submatrix, the first in the submatrix Line number The column element value is the i-th element in the projected coordinate matrix of the state feature matrix. The row vector and the standard projected coordinate matrix of the nth row vector The geodesic distance between each row vector. The arithmetic mean of all elements in the submatrix is ​​taken, and the result of the arithmetic mean is used as the manifold distance between the projected coordinate matrix of the state feature matrix and the standard projected coordinate matrix of the current pipe segment under the current operating condition.

[0042] In practice, all operating conditions corresponding to the current pipe segment are traversed. These conditions include normal operation, leakage, and blockage. The manifold distance values ​​between the projected coordinate matrix of the state feature matrix and the standard projected coordinate matrix of the normal operation, leakage, and blockage conditions are obtained. These three manifold distance values ​​are then sorted in ascending order (from smallest to largest). The manifold distance value with the smallest value is identified, and the operating condition corresponding to this smallest manifold distance value is marked as the preliminary operating condition for the current pipe segment.

[0043] The minimum manifold distance is compared to a preset distance threshold. The distance threshold is set to... The threshold value is set based on the following: calculating the statistical distribution of manifold distance for each pipe segment under the same operating condition using multiple batches of historical data, taking the average of the 95th percentile values ​​of the manifold distance for each pipe segment under the same operating condition, and then rounding down to two decimal places. If the minimum manifold distance value is less than or equal to the distance threshold value. If the initial working condition is determined, the safety status category of the detection unit will be output. If the minimum manifold distance value is greater than the distance threshold value... If so, the safety status category of the detection unit will be marked as an unknown operating condition category and output.

[0044] In practice, the safety status category output results of each detection unit are obtained sequentially along the pipe segment topology order at the end of the gas pipeline network. The safety status categories of each detection unit are then arranged according to the pipe segment topology order to generate a safety status category sequence. The safety status category sequence is a one-dimensional sequence, where each element corresponds to a safety status category label for a detection unit. The values ​​of the safety status category labels are normal operating condition label, leakage operating condition label, blockage operating condition label, or unknown operating condition label.

[0045] The system performs continuous identification of all detection units marked as leaking conditions in the safety status category sequence. The continuous identification process is as follows: starting from the beginning of the safety status category sequence, the safety status category label of each element is read sequentially, and detection units that are consecutively adjacent and both have the safety status category label "leaking condition" are grouped together. A preset quantity threshold is set. The threshold is set based on the following: when the number of detection units for continuous leakage is less than... When the leakage is localized, it does not constitute a regional leakage event; when the number of detection units for continuous leakage is greater than or equal to This indicates that the leak has spread to multiple pipe sections along the gas delivery direction and needs to be treated as a regional leak. Detection units in the safety status category sequence that have been continuously identified and grouped together, with the number of detection units within the group exceeding a certain threshold, are considered appropriate. The continuous leakage condition detection unit set is merged into a leakage area.

[0046] The system summarizes the location information, the number of detection units within each leak area, and the safety status category of each detection unit after merging. The location information of a leak area consists of the starting and ending detection unit numbers, which correspond to the sequence numbers in the pipe segment topology. The system then summarizes the location information, the number of detection units, the safety status category of each detection unit, and the safety status category of detection units operating under non-leakage conditions to generate and output the gas pipeline network end-point safety inspection results.

[0047] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A gas pipeline network end-of-pipe safety detection system based on the Internet of Things, characterized in that, The system includes: The signal acquisition module collects pipe wall vibration signals and gas pressure fluctuation signals at each sensing node by using a piezoelectric sensor array deployed on the outer wall of the end section of the gas pipeline network. The feature extraction module extracts the vibration main frequency offset and vibration energy attenuation rate based on the pipe wall vibration signal at each sensing node, and extracts the pressure fluctuation phase difference and pressure pulsation amplitude ratio based on the gas pressure fluctuation signal inside the pipe. The vector construction module takes the pipe segment between two adjacent sensing nodes as the detection unit and uses the difference in vibration main frequency offset, vibration energy attenuation rate, pressure fluctuation phase difference, and pressure pulsation amplitude ratio between the sensing nodes at both ends of the detection unit to construct the multi-dimensional state feature vector of the detection unit. The matrix generation module arranges the multivariate state feature vectors of all detection units according to the pipe segment topology order to form the state feature matrix of the end of the gas pipeline network. The safety determination module maps the state feature matrix to a pre-constructed physical behavior feature space at the end of the gas pipeline network. In the physical behavior feature space, it calculates the manifold distance between the state feature matrix and the standard behavior pattern feature matrix of each pipe segment. Based on the manifold distance, it determines the safety status category of each detection unit and outputs the safety detection result.

2. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 1, characterized in that, The method involves collecting pipe wall vibration signals and gas pressure fluctuation signals at each sensing node using a piezoelectric sensor array deployed on the outer wall of the end section of the gas pipeline network. This includes: Along the gas delivery direction at the end of the gas pipeline network, piezoelectric accelerometers and piezoresistive pressure sensors are installed on the outer walls at both ends of each pipe section. A piezoelectric sensing array is constructed using all piezoelectric accelerometers and piezoresistive pressure sensors. A uniform sampling frequency and sampling duration are set to synchronously collect pipe wall vibration acceleration time history data and pipe gas pressure time history data at each sensing node. The pipe wall vibration acceleration time history data is integrated twice to obtain the pipe wall vibration displacement signal, and the gas pressure time history data in the pipe is digitally low-pass filtered to obtain the gas pressure fluctuation signal in the pipe.

3. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 2, characterized in that, The extraction of the vibration dominant frequency offset and vibration energy attenuation rate based on the pipe wall vibration signal at each sensing node includes: A short-time Fourier transform is performed on the pipe wall vibration displacement signal at each sensing node to obtain the vibration spectrum at each moment. The frequency component with the largest amplitude in the vibration spectrum is marked as the instantaneous vibration dominant frequency at that moment. Arrange the instantaneous dominant vibration frequencies of all moments in a time series, calculate the slope of the linear trend of the time series, and use the slope of the linear trend as the offset of the dominant vibration frequency. The time-domain energy envelope of the pipe wall vibration displacement signal at each sensing node is extracted to obtain the exponential decay coefficient of the vibration signal amplitude envelope, and this exponential decay coefficient is used as the vibration energy decay rate.

4. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 2, characterized in that, The step of extracting the pressure fluctuation phase difference and pressure pulsation amplitude ratio based on the gas pressure fluctuation signal in the pipe includes: Bandpass filtering is performed on the gas pressure fluctuation signal in the pipe collected at the same sensing node to extract the pressure fluctuation component within a preset frequency band. Perform a Hilbert transform on the pressure fluctuation component to obtain its instantaneous phase sequence and instantaneous amplitude sequence; Linear fitting is performed on the instantaneous phase sequence along the time axis, and the slope of the fitted line is used as the pressure fluctuation phase difference at the sensing node. The ratio of the time average of the instantaneous amplitude sequence to the static pressure of the gas in the pipe at the sensing node is taken as the pressure pulsation amplitude ratio.

5. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 1, characterized in that, The method uses the pipe segment between two adjacent sensing nodes as the detection unit. It constructs a multivariate state feature vector for the detection unit using the differences in vibration dominant frequency offset, vibration energy attenuation rate, pressure fluctuation phase difference, and pressure pulsation amplitude ratio between the sensing nodes at both ends of the detection unit. This vector includes: The difference in the main frequency offset of the detection unit is obtained by subtracting the vibration main frequency offset of the upstream sensing node from the vibration main frequency offset of the downstream sensing node. The energy attenuation rate of the upstream sensing node of the detection unit is subtracted from the vibration energy attenuation rate of the downstream sensing node to obtain the energy attenuation rate difference of the detection unit. The phase difference of the pressure fluctuation at the upstream sensing node of the detection unit is obtained by subtracting the phase difference of the pressure fluctuation at the downstream sensing node. The difference in amplitude ratio of the upstream sensor node to the downstream sensor node of the detection unit is obtained by subtracting the amplitude ratio of the upstream sensor node from the amplitude ratio of the downstream sensor node. The difference in main frequency offset, energy attenuation rate, phase difference, and amplitude ratio are combined in a fixed-dimensional order to form the multivariate state feature vector of the detection unit.

6. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 1, characterized in that, The process of arranging the multivariate state feature vectors of all detection units according to the pipe segment topology to form a state feature matrix at the end of the gas pipeline network includes: According to the actual flow direction of the gas, obtain the sequential order of all detection units at the end of the gas pipeline network; The initial state feature matrix is ​​constructed by using the multivariate state feature vector of each detection unit as a row of the matrix, the order of the detection units as the row index, and the dimension components of the multivariate state feature vector as the column index. Each element in the initial state feature matrix is ​​subjected to deviation standardization, so that the numerical range of each element is mapped to the interval [0,1]. The matrix after deviation standardization is used as the state feature matrix.

7. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 1, characterized in that, The step of mapping the state feature matrix to a pre-constructed physical behavior feature space at the end of the gas pipeline network includes: Obtain the physical property parameters of each pipe segment at the end of the gas pipeline network, including pipe segment length, pipe segment inner diameter, pipe segment wall thickness, and number of pipe segment bends; Construct a physical attribute vector for each pipe segment using the pipe segment length, pipe segment inner diameter, pipe segment wall thickness, and number of pipe segment elbows; Using the physical attribute vectors of all pipe segments as basis vectors, a linear transformation projection is performed on the state feature matrix to map the state feature matrix from the original feature dimension space to the physical behavior feature space spanned by the basis vectors, thereby obtaining the projection coordinate matrix of the state feature matrix in the physical behavior feature space.

8. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 7, characterized in that, The calculation of the manifold distance between the state feature matrix and the standard behavior pattern feature matrix of each pipe segment in the physical behavior feature space includes: The standard state feature matrix corresponding to each pipe section under normal operating conditions, leakage operating conditions and blockage operating conditions is extracted from the historical database, and each standard state feature matrix is ​​mapped to the physical behavior feature space to obtain the standard projection coordinate matrix corresponding to each pipe section under each operating condition. In the physical behavior feature space, the geodesic distance between the projected coordinate matrix of the state feature matrix and the standard projected coordinate matrix of each working condition of the current pipe segment is calculated, and this geodesic distance is used as the manifold distance.

9. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 8, characterized in that, The determination of the safety status category of each detection unit based on manifold distance includes: Traverse all operating conditions corresponding to the current pipe segment and obtain the manifold distance values ​​between the projected coordinate matrix of the state feature matrix and the standard projected coordinate matrix of each operating condition. All manifold distance values ​​are sorted in ascending order, and the working condition corresponding to the minimum manifold distance value is marked as the preliminary judgment working condition for the current pipe segment. The minimum manifold distance value is compared with a preset distance threshold. If the minimum manifold distance value is less than or equal to the distance threshold, the preliminary judgment condition is output as the safety status category of the detection unit.

10. The gas pipeline network end-of-line safety detection system based on the Internet of Things as described in claim 1, characterized in that, The process of determining the safety status category of each detection unit based on manifold distance and outputting the safety detection result includes: Following the topological order of the pipe segment, the safety status category of each detection unit is obtained sequentially, generating a safety status category sequence; The detection units marked as leakage conditions in the safety status category sequence are continuously identified, and the leakage condition detection units that are continuously arranged and exceed the preset number threshold are merged into a leakage area. The location information of the leak area, the number of detection units contained in the leak area, and the safety status category of each detection unit are summarized and output as the safety detection result of the gas pipeline network end.

Images