Intelligent detection and early warning positioning method for oil and gas pipeline leakage

Abstract

The application discloses an oil and gas pipeline leakage intelligent detection and early warning positioning method, which comprises the following steps: acquiring pipeline vibration wave, pressure change and temperature field signals along the line through multi-dimensional signal cooperative acquisition, inputting the signals into a spectral space attention detection network for spatial and spectral dimension joint feature mining, and generating a high-dimensional leakage feature vector; calling a ground penetrating radar signal inversion model to analyze the change of medium electromagnetic parameters around the pipeline, positioning the preliminary range of the potential leakage area, and simultaneously establishing a pressure propagation and distance mapping relationship by using a pressure curve distance positioning algorithm; inputting the inversion result and the positioning result into a pipeline leakage intelligent early warning decision platform for multi-source data fusion verification, outputting accurate leakage position coordinates and leakage degree characteristic parameters, and forming a complete technical link from signal acquisition, feature mining, inversion positioning to fusion decision, so that the accuracy of leakage detection and the timeliness of early warning response are improved.

Description

Technical Field

[0001] This invention relates to the field of oil and gas pipeline leakage detection technology, and in particular to an intelligent detection, early warning and location method for oil and gas pipeline leakage. Background Technology

[0002] Oil and gas pipelines, as core infrastructure for energy transportation, are widely distributed in complex environments such as land and sea. Their operational safety is directly related to the stability of energy supply and the safety of the ecological environment. During long-term service, pipelines are affected by multiple factors such as geological subsidence, corrosion and aging, and third-party damage, resulting in a persistent risk of leakage. Leakage accidents not only cause significant energy losses but may also lead to serious consequences such as fires, explosions, and environmental pollution. With the continuous increase in pipeline mileage, the diversification of transported media, and the increasing complexity of operating conditions, traditional detection methods are no longer sufficient to meet the demands for real-time and accurate early warning. There is an urgent need to build an efficient and integrated intelligent detection and positioning technology system to achieve early identification, rapid location, and timely warning of potential leaks, providing technical support for the safe operation and maintenance of pipelines.

[0003] Existing technologies for detecting and locating leaks in oil and gas pipelines have two key drawbacks: First, the synergy between feature extraction and signal analysis is insufficient, relying heavily on single-dimensional signal data for detection. This fails to fully exploit the correlation features between different signals in spatial and spectral dimensions, resulting in limited ability to identify weak leak signals in complex environments and susceptibility to interference signals, leading to misjudgments. Second, the accuracy of location and the efficiency of data fusion need improvement. Existing methods do not tightly integrate the inversion model with the location algorithm, lacking a systematic verification mechanism for multi-source detection results. This makes it difficult to accurately analyze the changes in medium parameters and pressure propagation patterns in the leak area, resulting in significant deviations in leak location. Furthermore, the lack of effective integration of technical parameters from various stages during the early warning decision-making process affects the timeliness and reliability of the early warning response. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides an intelligent detection, early warning and location method for oil and gas pipeline leaks.

[0005] The technical solution adopted in this invention is an intelligent detection, early warning, and location method for oil and gas pipeline leaks, comprising the following steps: S1, collecting vibration wave signals, medium pressure change signals, and pipeline surface temperature field signals along the oil and gas pipeline using distributed sensing units to construct a multi-dimensional raw signal dataset; S2, inputting the multi-dimensional raw signal dataset into a spectral spatial attention detection network, and performing joint feature mining of the spatial and spectral dimensions of the signals through the network's feature extraction layer to generate a high-dimensional leak feature vector; S3, calling a ground-penetrating radar signal inversion model to invert and calculate the abnormal medium response features in the high-dimensional leak feature vector, and analyzing the medium around the pipeline. The process involves: S4, analyzing the electromagnetic parameter variation patterns to initially locate the potential leak area; S5, using a pressure curve distance positioning algorithm to perform spatiotemporal correlation analysis on the characteristic parameters corresponding to the medium pressure change signal, establishing a pressure propagation model and distance mapping relationship; S6, inputting the ground-penetrating radar signal inversion results and pressure curve distance positioning results into the pipeline leak intelligent early warning decision platform for multi-source data fusion verification, outputting accurate leak location coordinates and leak degree characteristic parameters; and S7, synchronizing the leak location coordinates, leak degree characteristic parameters, and early warning level information to the terminal monitoring system through the signal transmission module of the pipeline leak intelligent early warning decision platform, forming a closed-loop detection and early warning process.

[0006] Furthermore, the feature extraction expression of the spectral spatial attention detection network is: ,in, To leak high-dimensional feature vectors, It is the Sigmoid activation function. These are the spectral dimension weighting coefficients. for Convolution operation, The original multi-dimensional signal dataset, For element-wise multiplication, The spatial dimension weight coefficients, Attention ( ) is the spatial attention computation function. For spatial attention adjustment parameters, For feature extraction bias term, For batch normalization operations, This is a max pooling operation.

[0007] Furthermore, the inversion calculation expression of the ground-penetrating radar signal inversion model is as follows: ,in, These are characteristic parameters of the abnormal response of the medium. The length of the pipeline inspection section. The relative permittivity of the medium surrounding the pipe. The relative permeability of the medium surrounding the pipe. This is the frequency-time domain data of the raw ground-penetrating radar signal. To detect planar coordinates, The reference coordinates for the ground-penetrating radar antenna. The signal attenuation coefficient is... For radar signal frequency, For the duration of transmission.

[0008] Furthermore, the distance calculation expression for the pressure curve distance positioning algorithm is as follows: ,in, The distance between the leak point and the detection point. The number of pressure detection points. For the first Each testing point Changes in pressure over time. For the first The medium flow velocity at each detection point For the first The angle between the pressure wave propagation direction at each detection point and the pipeline axis. for Pipeline position at all times Pressure value, For time intervals, This is the distance correction factor. The pressure change threshold, It is a symbolic function.

[0009] Furthermore, the multi-source data fusion expression of the intelligent early warning decision-making platform for pipeline leaks is as follows: ,in, To integrate decision-making outputs, As the weights of the ground-penetrating radar inversion results, Weight the pressure location results. For the feature vector weights, For the fusion adjustment coefficient, for Convolution operation, This is a feature splicing operation.

[0010] Furthermore, the expression for determining the warning level of the intelligent early warning decision-making platform for pipeline leaks is as follows: ,in, At the warning level, It is the L2 norm. For the weighting coefficients of the decision outcome, The weighting coefficients are the mean of the eigenvectors. Thresholds are set for the warning level. This is a floor operation.

[0011] Further, step S3 includes the following sub-steps: S31, filtering the electromagnetic response-related features in the high-dimensional leakage feature vector, extracting a subset of features related to the changes in dielectric constant and magnetic permeability of the medium surrounding the pipeline, and establishing a mapping relationship between features and medium parameters; S32, inputting the feature subset into the initialization module of the ground-penetrating radar signal inversion model, setting the spatial grid division parameters and iterative convergence conditions for the inversion calculation, and determining the initial boundary values ​​for the inversion calculation; S33, spatially discretizing the detection area according to the set grid parameters, and simulating the propagation path and signal attenuation law of radar waves under different medium parameters through the forward modeling module of the model; S34, comparing the forward modeling simulation results with the actual collected radar signal features, and iteratively optimizing and adjusting the estimated values ​​of medium parameters until the consistency between the inversion results and the actual signal meets the set requirements.

[0012] Further, step S4 includes the following sub-steps: S41, extracting pressure time series data of each detection point from the multi-dimensional original signal dataset, removing trend components from the signal, and retaining dynamic pressure change components related to leakage; S42, extracting time-domain features from the processed pressure time series data, obtaining pressure peak value, rising slope, and fluctuation frequency calibration feature parameters, and constructing a pressure feature matrix; S43, inputting the pressure feature matrix into the pressure curve distance positioning algorithm, and combining pipeline material parameters and medium physical property parameters to establish a calculation model for pressure wave propagation speed and distance; S44, using the algorithm's spatiotemporal matching module to synchronize and spatially correlate the pressure features of different detection points, calculating the distance estimate from each detection point to the potential leakage point, and forming a distance candidate set.

[0013] Further, S5 includes the following sub-steps: S51, converting the potential leak area range data obtained from ground-penetrating radar signal inversion into regional boundary parameters under a unified spatial coordinate system, and spatially matching them with the distance candidate set obtained from pressure curve distance positioning; S52, calling the multi-source data fusion module of the pipeline leak intelligent early warning decision platform, and using a weighted fusion strategy to fuse the inversion results, positioning results, and high-dimensional leak feature vectors to generate a comprehensive feature matrix; S53, analyzing the comprehensive feature matrix through the platform's decision judgment module, and filtering out target areas that meet the leak characteristics according to preset leak judgment rules to determine the precise leak location coordinates; S54, calculating the leak degree feature parameters based on the degree of change of medium parameters and pressure loss corresponding to the precise leak location coordinates to form a complete detection and positioning result.

[0014] A smart detection and early warning method for oil and gas pipeline leaks is proposed. This method is implemented through different units, including: a multi-dimensional signal collaborative acquisition unit, a spectral spatial attention feature extraction unit, a ground-penetrating radar signal inversion and calculation unit, a pressure curve distance positioning analysis unit, a multi-source data fusion decision unit, and a detection and early warning information transmission unit. The multi-dimensional signal collaborative acquisition unit and the spectral spatial attention feature extraction unit are connected through a high-speed data transmission interface to transmit the acquired multi-dimensional raw signal dataset to the feature extraction unit in real time. The spectral spatial attention feature extraction unit establishes bidirectional data interaction channels with the ground-penetrating radar signal inversion and calculation unit and the pressure curve distance positioning analysis unit, respectively, to generate high-dimensional leak features. Vector synchronization is distributed to two computing units; the ground-penetrating radar signal inversion computing unit and the pressure curve distance positioning analysis unit are both connected to the multi-source data fusion decision unit through a data bus, transmitting their respective calculation results to the fusion decision unit for data fusion processing; the multi-source data fusion decision unit is connected to the detection and early warning information transmission unit through a communication protocol interface, transmitting the final leak location coordinates, leak degree characteristic parameters, and early warning level information to the transmission unit; the detection and early warning information transmission unit establishes a connection with the terminal monitoring system through a wireless communication network to push and provide feedback on detection and early warning information in real time. All units work collaboratively through a distributed control module to form a complete technical link from signal acquisition to early warning output.

[0015] Beneficial Effects: This invention proposes an intelligent detection, early warning, and location method for oil and gas pipeline leaks. By combining multi-dimensional signal collaborative acquisition with a spectral spatial attention detection network, it systematically mines the correlation features of signals in the spatial and spectral dimensions, breaking through the limitations of traditional single-dimensional signal detection. This significantly improves the ability to identify weak leak signals in complex environments and effectively solves the problems of insufficient synergy between feature extraction and signal analysis in existing technologies, as well as susceptibility to misjudgment by interference signals. Furthermore, by utilizing the close linkage between a ground-penetrating radar signal inversion model and a pressure curve distance positioning algorithm, combined with the multi-source data fusion and verification mechanism of a pipeline leak intelligent early warning decision platform, it accurately analyzes leaks. By studying regional media parameter changes and pressure propagation patterns, this invention significantly reduces leak location deviation. Simultaneously, through the coordinated operation and parameter integration of various technical modules, it improves the response efficiency of early warning decisions, overcoming the shortcomings of existing technologies in terms of positioning accuracy and data fusion efficiency. This invention achieves a closed-loop design across the entire process, from signal acquisition, feature extraction, inversion calculation, distance positioning to fusion decision-making and information transmission. This ensures both the comprehensiveness and accuracy of leak detection and enhances the timeliness and reliability of early warning and location, providing efficient technical support for the safe operation and maintenance of oil and gas pipelines and meeting the practical needs for early identification, rapid location, and timely warning of pipeline leaks under complex operating conditions. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method steps of the present invention;

[0017] Figure 2 This is a diagram showing the unit composition of the method implementation of the present invention;

[0018] Figure 3 A system field control diagram constructed for the method of this invention;

[0019] Figure 4 This is a diagram of the backend code interface for implementing the method of the present invention. Detailed Implementation

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] like Figure 1 As shown, the intelligent detection and early warning location method for oil and gas pipeline leaks includes the following steps: S1, collecting vibration wave signals, medium pressure change signals, and pipeline surface temperature field signals along the oil and gas pipeline using distributed sensing units to construct a multi-dimensional raw signal dataset; S2, inputting the multi-dimensional raw signal dataset into a spectral spatial attention detection network, and performing joint feature mining of the spatial and spectral dimensions of the signals through the network's feature extraction layer to generate a high-dimensional leak feature vector; S3, calling a ground-penetrating radar signal inversion model to invert and calculate the medium anomaly response features in the high-dimensional leak feature vector, and analyzing the electromagnetic parameters of the medium surrounding the pipeline. S4. Using the pressure curve distance positioning algorithm, the characteristic parameters corresponding to the medium pressure change signal are analyzed in time and space to establish a pressure propagation model and distance mapping relationship; S5. The ground penetrating radar signal inversion results and pressure curve distance positioning results are input into the pipeline leakage intelligent early warning decision platform for multi-source data fusion verification, and the precise leakage location coordinates and leakage degree characteristic parameters are output; S6. The leakage location coordinates, leakage degree characteristic parameters and early warning level information are synchronized to the terminal monitoring system through the signal transmission module of the pipeline leakage intelligent early warning decision platform to form a closed-loop detection and early warning process.

[0022] In step S1, a distributed sensing unit is deployed every 30 meters along the oil and gas pipeline. Each unit integrates a vibration wave sensor, a pressure sensor, and a temperature sensor. The sensor sampling frequency is set to 1000 Hz, and the data acquisition cycle is 2 seconds. The sensing units are connected to the data acquisition terminal via fiber optic links. The vibration wave sensor captures vibration signals on the pipeline surface in the range of 0.1 to 100 Hz. The pressure sensor monitors dynamic data of pressure changes in the pipeline medium in the range of 0 to 10 MPa. The temperature sensor collects temperature field distribution information on the pipeline surface from -20 to 80 degrees Celsius. During the acquisition process, a synchronization clock module is used to achieve time synchronization of all sensing units, with the time synchronization error controlled within 1 millisecond to ensure the consistency of multi-dimensional signals in the time dimension. The collected vibration wave signals, medium pressure change signals, and pipeline surface temperature field signals are stored in binary format on the local hard drive of the data acquisition terminal, with a storage capacity of no less than 1TB. At the same time, the real-time collected data is backed up to the cloud server through a 5G communication module, forming a multi-dimensional raw signal dataset covering the entire pipeline mileage. This provides comprehensive and accurate raw data support for subsequent feature extraction and leak detection. This step ensures the ability to capture minute abnormal signals in the pipeline through high-density deployment of sensing units and high-frequency data acquisition, avoiding signal omissions caused by excessively large sampling intervals or sparse distribution of sensing units.

[0023] In step S2, the multi-dimensional raw signal dataset is first downloaded from the cloud server to the data processing terminal. The data processing terminal is configured with at least an 8-core CPU and 16GB of memory to ensure data processing efficiency. The downloaded multi-dimensional raw signal dataset is then categorized and organized according to time series and spatial location, forming a signal data matrix indexed by pipeline mileage coordinates. This data matrix is ​​then input into a pre-trained spectral spatial attention detection network. This network includes an input layer, a feature extraction layer, an attention mechanism layer, and an output layer. The feature extraction layer consists of 8 convolutional blocks, each containing 64 3×3 convolutional kernels. The convolution stride is set to 1, and the padding method uses same padding. Spatial dimension features are extracted from the signal through convolution operations. Simultaneously, the network decomposes the frequency components of the signal through a spectral dimension analysis module, extracting spectral features in three frequency bands: 0.1 to 10 Hz, 10 to 50 Hz, and 50 to 100 Hz. The attention mechanism layer calculates the correlation weights between spatial and spectral features, assigning high weights (0.8 to 1.0) to important features and low weights (0.1 to 0.3) to interfering features, thus achieving adaptive feature selection and enhancement. Finally, the fully connected network in the output layer maps the selected high-dimensional features into a high-dimensional leakage feature vector of length 1024. This vector includes the signal's spatial distribution features, spectral frequency features, and cross-dimensional correlation features, providing a highly discriminative feature foundation for subsequent inversion calculations and localization analysis. This step, through multi-level feature mining and precise selection via the attention mechanism, significantly improves the distinguishability between leakage and interfering features.

[0024] In step S3, feature components related to the electromagnetic response of the surrounding medium are first extracted from the high-dimensional leakage feature vector. These feature components include signal features corresponding to changes in the medium's dielectric constant and permeability. The extraction process is completed using a feature filter, whose feature thresholds are pre-calibrated based on the pipe material and the type of the surrounding medium. The filtered feature components are then input into the ground-penetrating radar signal inversion model. During model initialization, spatial grid partitioning parameters are set, dividing the detection area into a 100×100×50 three-dimensional grid with a grid step size of 0.1 meters. The iteration convergence condition is set to ensure the error between two iterations is less than 0.001. During the inversion calculation, the model, based on electromagnetic propagation theory, calculates the propagation path and signal attenuation of radar waves under different medium parameters through forward modeling. The forward modeling simulation uses the finite difference method, with a time step of 1 nanosecond and at least 100 simulations. The forward simulation results were compared with the actual acquired radar signal characteristics. The estimated values ​​of the medium parameters were adjusted using a gradient descent algorithm, with an adjustment step size of 0.01 for each iteration, and the number of iterations was controlled between 50 and 100, until the consistency between the inversion results and the actual signal characteristics reached more than 95%. The distribution patterns of the relative permittivity of the medium around the pipeline, ranging from 1 to 10, and the relative permeability, ranging from 1 to 2, were obtained through inversion calculation and analysis. Based on the spatial distribution range of abnormal changes in medium parameters, the preliminary boundary of the potential leak area was located. The spatial error of the preliminary range was controlled within 1 meter. This step, through fine-grained grid division and iterative optimization, ensured the accuracy of the potential leak area location, laying the foundation for subsequent precise location.

[0025] In step S4, the pressure change signals of the medium at each pressure detection point are first extracted from the multi-dimensional original signal dataset. Pressure data from 1000 consecutive sampling points are extracted for each detection point to form a pressure time series. The pressure time series is then processed to remove the trend term, retaining the dynamic pressure change component. The processing uses a moving average method with a sliding window size of 50 sampling points. Subsequently, time-domain feature extraction is performed on the processed pressure time series to obtain key feature parameters such as pressure peak value, rising slope, fluctuation frequency, and pressure change duration. The pressure peak value is determined by traversing the time series data; the rising slope is obtained by linear fitting to calculate the pressure change rate of 50 adjacent sampling points; the fluctuation frequency is obtained through fast Fourier transform analysis; and the pressure change duration is determined by counting the number of sampling points where the pressure value exceeds the normal fluctuation range. The extracted key feature parameters are input into the pressure curve distance positioning algorithm. The algorithm combines the elastic modulus of the pipe material and the density and viscosity of the medium, which are pre-entered into the algorithm database based on the pipe design documents and medium testing reports. The algorithm establishes a pressure wave propagation velocity model to calculate the propagation velocity of pressure waves within the pipeline. The propagation velocity calculation considers the influence of pipeline diameter, wall thickness, and medium temperature, with correction coefficients set between 0.95 and 1.05. Based on the time difference between the pressure wave propagation velocity and the pressure signal at each detection point, a mapping relationship between pressure propagation and distance is established. The estimated distance from each detection point to the potential leak point is calculated, with the accuracy of the distance estimation controlled within 0.5 meters. This step, through multi-parameter fusion and refined calculation, achieves preliminary location of the leak distance.

[0026] In step S5, the data format of the ground-penetrating radar signal inversion results and the pressure curve distance positioning results are first standardized. The boundary parameters of the potential leakage area obtained from the inversion are converted into a unified geodetic coordinate system. The coordinate transformation adopts Gaussian projection transformation, and the transformation error is controlled within 0.1 meters. The distance estimates obtained from the pressure curve distance positioning are converted into candidate location points within the coordinate range, with 5 candidate location points generated for each detection point. Subsequently, the standardized two types of data are input into the intelligent early warning decision-making platform for pipeline leakage. The platform adopts a distributed computing architecture, including a data receiving module, a fusion verification module, a decision analysis module, and a result output module. The data receiving module receives the two types of data through the TCP / IP protocol, with a receiving bandwidth of no less than 100Mbps to ensure real-time data transmission. The fusion verification module adopts a weighted fusion strategy, assigning a weight of 0.6 to the ground-penetrating radar inversion results and a weight of 0.4 to the pressure positioning results. The spatial location information of the two types of data is fused by calculating the weighted average value. At the same time, the leakage degree-related features in the high-dimensional leakage feature vector are combined to verify the fusion result. The verification process is achieved by comparing whether the changes in the medium parameters corresponding to the fused location points match the pressure loss. Based on the fusion verification results and according to preset leak location determination rules, the decision analysis module selects precise leak location coordinates that match the leak characteristics, with coordinate accuracy controlled within 0.3 meters. Simultaneously, based on the change range of media parameters corresponding to the leak location, pressure loss rate, and the anomaly degree of the high-dimensional feature vector, it calculates leak severity characteristic parameters. These parameters are divided into 10 levels, determined based on the media leakage amount and pipeline operational safety thresholds. Finally, the results output module outputs the precise leak location coordinates and leak severity characteristic parameters with an output delay of no more than 1 second. This step, through multi-source data fusion and rigorous verification, ensures the accuracy and reliability of the leak detection and location results.

[0027] During step S6, the signal transmission module of the intelligent early warning decision-making platform for pipeline leaks employs a combination of wired and wireless transmission methods. Wired transmission is achieved through fiber optic communication links, while wireless transmission utilizes 5G and LoRa dual-mode communication to ensure signal coverage in complex environments. The transmission module first encapsulates the leak location coordinates, leak severity characteristic parameters, and early warning level information using JSON format and a data compression ratio of 2:1 to reduce the amount of data transmitted. The encapsulated data is then transmitted via a wired link to the terminal monitoring system at the pipeline operation and maintenance center, with a transmission rate of no less than 1Gbps. Simultaneously, the early warning information is pushed to the mobile terminals of maintenance personnel via a wireless communication module. The mobile terminals support both Android and iOS systems, with an information push delay of no more than 3 seconds. The terminal monitoring system includes a data parsing module, a visualization module, and a feedback module. The data parsing module decompresses and parses the received encapsulated data, achieving a 100% accuracy rate. The visualization module visually displays the leak location coordinates on an electronic map, annotating leak severity characteristics and warning levels. The map zoom scale supports adjustment from 1:100 to 1:10000, and also displays the leak severity trend over time in chart form. The feedback module transmits the terminal's received confirmation information back to the pipeline leak intelligent early warning decision platform, forming a closed-loop detection and early warning process. This step, through multi-channel transmission and terminal visualization, ensures that early warning information is delivered to relevant personnel in a timely manner, providing rapid response support for leak emergency handling. Simultaneously, the closed-loop process design guarantees the integrity and traceability of the detection and early warning system.

[0028] Preferably, the feature extraction expression of the spectral spatial attention detection network is: ,in, To leak high-dimensional feature vectors, It is the Sigmoid activation function. These are the spectral dimension weighting coefficients. for Convolution operation, The original multi-dimensional signal dataset, For element-wise multiplication, The spatial dimension weight coefficients, Attention ( ) is the spatial attention computation function. For spatial attention adjustment parameters, For feature extraction bias term, For batch normalization operations, This is a max pooling operation.

[0029] Specifically, the spectral spatial attention detection network extracts features based on the synergistic correlation between the spatial distribution characteristics and spectral frequency characteristics of the signal. It mines local spatial features through convolution operations, strengthens the weights of key features using an attention mechanism, and further enhances feature stability by incorporating batch normalization and pooling operations. This formula leverages the spatial clustering and spectral frequency specificity of multi-dimensional signals in a leaky state, element-wise fusing the spatial features extracted by convolution with the weight coefficients calculated by attention. After adding a bias term, an activation function enhances the nonlinear expression, and finally, a product operation is performed with the pooled features to achieve deep coupling between spatial and spectral features. The parameter values ​​have been calibrated through extensive experiments, with both spectral and spatial dimension weight coefficients ranging from 0.5 to 1.2, spatial attention adjustment parameters ranging from 0.1 to 0.3, and feature extraction bias terms set from 0.01 to 0.05. In practice, the original multi-dimensional signal is first input into a 3×3 convolutional layer for spatial feature extraction. The attention calculation module analyzes the correlation strength between the spatial distribution of the signal and the spectral frequency band and assigns weights. After activation function processing, it is fused with the max pooled features. Then, batch normalization is used to eliminate data distribution differences, and finally, a high-dimensional leakage feature vector is output. This formula effectively improves the distinguishability between leakage features and interference features through the synergistic effect of multiple modules, providing accurate feature input for subsequent inversion calculations.

[0030] Preferably, the inversion calculation expression of the ground-penetrating radar signal inversion model is: ,in, These are characteristic parameters of the abnormal response of the medium. The length of the pipeline inspection section. The relative permittivity of the medium surrounding the pipe. The relative permeability of the medium surrounding the pipe. This is the frequency-time domain data of the raw ground-penetrating radar signal. To detect planar coordinates, The reference coordinates for the ground-penetrating radar antenna. The signal attenuation coefficient is... For radar signal frequency, For the duration of transmission.

[0031] Specifically, the ground-penetrating radar (GPR) signal inversion model expression is based on electromagnetic propagation theory and integral transform principle. It considers the propagation attenuation law of radar waves in the medium surrounding the pipeline and calculates the mapping relationship between the medium's electromagnetic parameters and the radar signal through double integral. This formula utilizes the abnormal changes in the dielectric constant and permeability of the medium surrounding the pipeline during a leak, which alter the radar wave propagation path and attenuation characteristics. The medium's electromagnetic parameters and the radar signal's frequency-time domain data are used as the integration kernel. The reciprocal of the propagation distance is used to correct for signal attenuation, and an exponential function characterizes the attenuation effect of frequency and time. The signal attenuation coefficient is set to 0.001 to 0.01, the radar signal frequency range is 100 to 500 MHz, the propagation time is 0 to 10 microseconds, the pipeline detection section length is set to 100 to 500 meters based on the actual pipeline mileage, and the GPR antenna reference coordinates are determined according to the detection layout location. During implementation, the detection area is first spatially discretized to obtain the initial values ​​of the medium electromagnetic parameters at each grid point. These values ​​are then substituted into the formula for integration to obtain the forward modeling simulation signal. This signal is compared with the actual acquired radar signal. The medium electromagnetic parameters are iteratively adjusted until the inversion result matches the actual signal to the required degree. This formula accurately characterizes the correlation between electromagnetic parameters and radar signals, enabling efficient analysis of the medium's abnormal response characteristics.

[0032] Preferably, the distance calculation expression of the pressure curve distance positioning algorithm is: ,in, The distance between the leak point and the detection point. The number of pressure detection points. For the first Each testing point Changes in pressure over time. For the first The medium flow velocity at each detection point For the first The angle between the pressure wave propagation direction at each detection point and the pipeline axis. for Pipeline position at all times Pressure value, For time intervals, This is the distance correction factor. The pressure change threshold, It is a symbolic function.

[0033] Specifically, the pressure curve distance positioning algorithm is based on pressure wave propagation theory and spatiotemporal correlation analysis. It combines parameters such as pressure change, medium flow velocity, and pressure wave propagation direction, establishing a distance mapping relationship through statistical averaging and partial derivative calculations. This formula utilizes the fact that pressure changes caused by leakage propagate along the pipeline as pressure waves. The propagation distance is positively correlated with the pressure change and flow velocity, and negatively correlated with the cosine of the angle between the pressure wave propagation direction and the pipeline axis. Simultaneously, the product of the pressure's spatiotemporal partial derivatives corrects for the influence of propagation velocity fluctuations, and a distance correction coefficient and sign function are introduced to optimize positioning accuracy. The pressure change threshold is set to 0.05 to 0.2 MPa, the distance correction coefficient is set to 0.8 to 1.1, the number of pressure detection points is set to 20 to 50 based on the pipeline length, the time interval is set to 0.1 to 0.5 seconds, and the angle between the pressure wave propagation direction and the pipeline axis ranges from 0 to 30 degrees. During implementation, the pressure change and medium flow velocity at each detection point are first extracted, the cosine value of the angle between the pressure wave propagation direction is calculated, the rate of change of pressure in the spatial and temporal dimensions is calculated through partial derivatives, and the results are substituted into the formula for summation and averaging. The sign function is then used to determine whether the pressure change exceeds the threshold, and finally the distance between the leak point and the detection point is obtained. This formula achieves accurate estimation of the leak distance through comprehensive calculation of multiple parameters.

[0034] Preferably, the multi-source data fusion expression of the intelligent early warning decision-making platform for pipeline leaks is: ,in, To integrate decision-making outputs, As the weights of the ground-penetrating radar inversion results, Weight the pressure location results. For the feature vector weights, For the fusion adjustment coefficient, for Convolution operation, This is a feature splicing operation.

[0035] Specifically, the multi-source data fusion expression of the intelligent early warning decision-making platform for pipeline leaks is based on weighted fusion theory and the principle of feature convolution. It considers the different importance of ground-penetrating radar inversion results, pressure location results, and high-dimensional feature vectors, achieving complementary fusion of multi-source data through weighted summation and convolutional convolution. This formula addresses the limitations of single detection results by highlighting the role of key data through weight allocation. Simultaneously, it improves the reliability of the decision output by performing dimensional transformation and feature fusion on the concatenated multi-source features using 1×1 convolution. The parameter values ​​were determined through multi-source data fusion experiments. The weights for ground-penetrating radar inversion results were set to 0.5 to 0.7, pressure location results to 0.3 to 0.5, and high-dimensional feature vectors to 0.4 to 0.6. The fusion adjustment coefficient ranged from 0.8 to 1.2. In practice, the medium anomaly response characteristic parameters obtained from ground-penetrating radar inversion, the distance parameters obtained from pressure curve positioning, and the high-dimensional leakage feature vector are first standardized. They are then weighted and summed according to set weights. The three types of data are then concatenated and input into a 1×1 convolutional layer for dimensionality compression and feature fusion. After superimposing the fusion adjustment coefficient, the fusion decision output is obtained. This formula effectively reduces the impact of single data errors and improves the accuracy of leakage detection decisions through deep fusion of multi-source data.

[0036] Preferably, the expression for determining the warning level of the intelligent early warning decision-making platform for pipeline leakage is: ,in, At the warning level, It is the L2 norm. For the weighting coefficients of the decision outcome, The weighting coefficients are the mean of the eigenvectors. Thresholds are set for the warning level. This is a floor operation.

[0037] Specifically, the intelligent early warning decision-making platform for pipeline leaks uses an expression based on norm calculation and mean statistics. It combines the strength of the fused decision results with the mean of the high-dimensional feature vector, and quantifies the early warning level through threshold division. The formula utilizes the L2 norm of the fused decision results to characterize the comprehensive strength of the leak characteristics, and the mean of the high-dimensional feature vector reflects the degree of leak anomaly. The weighted sum of these two factors, compared to the division threshold, is rounded down to obtain the discrete early warning level. Parameter values ​​are determined based on pipeline operation safety requirements and the degree of leak hazard. The weighting coefficient for the decision results is set to 0.6 to 0.8, the weighting coefficient for the mean of the feature vector is 0.2 to 0.4, the early warning level division threshold is 1 to 3, and the early warning levels are divided into 1 to 5 levels. During implementation, the L2 norm of the fusion decision output is first calculated, the mean of the high-dimensional leakage feature vector is statistically analyzed, and then multiplied by the corresponding weight coefficients and summed. The summation result is then compared with the warning level classification threshold, and the integer warning level is obtained by rounding down. This formula achieves accurate determination of the warning level by quantifying the correlation between the fusion decision result and the mean of the features, providing clear emergency response basis for operation and maintenance personnel and ensuring timely response to leakage accidents.

[0038] Preferably, step S3 includes the following sub-steps: S31, filtering the electromagnetic response-related features in the high-dimensional leakage feature vector, extracting a subset of features related to the changes in dielectric constant and magnetic permeability of the medium surrounding the pipeline, and establishing a mapping relationship between the features and the medium parameters; S32, inputting the feature subset into the initialization module of the ground-penetrating radar signal inversion model, setting the spatial grid division parameters and iterative convergence conditions for the inversion calculation, and determining the initial boundary values ​​for the inversion calculation; S33, spatially discretizing the detection area according to the set grid parameters, and simulating the propagation path and signal attenuation law of radar waves under different medium parameters through the forward modeling module of the model; S34, comparing the forward modeling simulation results with the actual collected radar signal features, and iteratively optimizing and adjusting the estimated values ​​of the medium parameters until the consistency between the inversion results and the actual signal meets the set requirements.

[0039] Specifically, step S3 achieves ground-penetrating radar signal inversion through a step-by-step process. In S31, when selecting electromagnetic response-related features from the high-dimensional leakage feature vector, a feature correlation threshold of 0.85 is set. Only a subset of features with a correlation exceeding the threshold for changes in dielectric constant and permeability is retained. A one-to-one correspondence between the subset and the medium parameters is established through a feature mapping table to ensure the targeted nature of feature selection. During execution of S32, the feature subset is input into the inversion model initialization module. The spatial grid division parameters are set to a three-dimensional grid size of 100×100×50, a grid step size of 0.1 meters, and the iteration convergence condition is set to an absolute error of less than 0.001 between two iterations. The initial boundary values ​​are calibrated based on pipeline design parameters and historical medium data to ensure the rationality of the initial conditions for the inversion calculation. In S33, the detection area is spatially discretized according to the set grid parameters. Forward simulation is performed using the finite difference method, with a time step set to 1 nanosecond and a fixed number of simulations of 120. This comprehensively simulates the propagation path and attenuation law of radar waves under different medium parameters, providing sufficient data support for subsequent comparisons. During the implementation of S34, the estimated values ​​of medium parameters are adjusted using a gradient descent algorithm. The adjustment step size is 0.01 in each iteration, and the number of iterations is controlled within 80 until the inversion results match the actual radar signal characteristics by more than 95%. This step-by-step design achieves accurate analysis of the changing patterns of medium parameters through progressively refined operations, ensuring that the initial positioning accuracy of the potential leakage area is controlled within 1 meter, laying the foundation for subsequent accurate positioning.

[0040] Preferably, step S4 includes the following sub-steps: S41, extracting pressure time series data of each detection point from the multi-dimensional original signal dataset, removing the trend component from the signal, and retaining the dynamic pressure change component related to leakage; S42, performing time-domain feature extraction on the processed pressure time series data to obtain pressure peak value, rising slope, and fluctuation frequency calibration feature parameters, and constructing a pressure feature matrix; S43, inputting the pressure feature matrix into the pressure curve distance positioning algorithm, and combining the pipe material parameters and medium physical property parameters to establish a calculation model for pressure wave propagation speed and distance; S44, using the algorithm's spatiotemporal matching module to perform time synchronization and spatial correlation of the pressure features of different detection points, calculating the distance estimate from each detection point to the potential leakage point, and forming a distance candidate set.

[0041] Specifically, step S4 refines the pressure curve distance localization algorithm step by step. In S41, when extracting pressure time series data for each detection point from the multi-dimensional original signal dataset, 1000 consecutive sampling points are extracted for each detection point. The moving average method is used to remove the trend term, and the sliding window size is set to 50 sampling points, effectively separating the dynamic pressure change component from the irrelevant trend component. During S42, time-domain features are extracted from the processed time series. The pressure peak is obtained by traversing the sampling point data. The rising slope is calculated by linear fitting of 50 adjacent sampling points. The fluctuation frequency is determined by statistical analysis of the period of the time-domain signal. The duration of pressure change is calculated by counting the number of sampling points where the pressure value exceeds the normal fluctuation range (±0.05). Finally, a pressure feature matrix with a dimension of 50×4 (50 detection points, 4 types of key features) is constructed. In step S33, the feature matrix is ​​input into the positioning algorithm, and physical property parameters such as the elastic modulus of the pipe material, the density and viscosity of the medium are recorded to establish a pressure wave propagation velocity model. A temperature correction factor is introduced into the model to dynamically adjust the calculated propagation velocity value according to the real-time temperature of the medium. The correction factor ranges from 0.92 to 1.08. During the implementation of step S44, the spatiotemporal matching module of the algorithm uses a time synchronization algorithm to calibrate the signals of each detection point, with the time synchronization error controlled within 1 millisecond. The distance estimate from each detection point to the potential leak point is calculated through spatial correlation analysis, with the distance calculation accuracy controlled within 0.5 meters. This step-by-step design, through the orderly advancement of feature extraction, model building, and spatiotemporal matching, fully explores the positioning information in the pressure signal, ensures the accuracy of the distance mapping relationship, and provides reliable distance data support for leak point location.

[0042] Preferably, step S5 includes the following sub-steps: S51, converting the potential leak area range data obtained from ground-penetrating radar signal inversion into regional boundary parameters under a unified spatial coordinate system, and spatially matching them with the distance candidate set obtained from pressure curve distance positioning; S52, calling the multi-source data fusion module of the pipeline leak intelligent early warning decision platform, and using a weighted fusion strategy to perform data fusion on the inversion results, positioning results, and high-dimensional leak feature vectors to generate a comprehensive feature matrix; S53, analyzing the comprehensive feature matrix through the platform's decision judgment module, and filtering out target areas that meet the leak characteristics according to preset leak judgment rules to determine the precise leak location coordinates; S54, calculating the leak degree feature parameters based on the degree of change of medium parameters and pressure loss corresponding to the precise leak location coordinates to form a complete detection and positioning result.

[0043] Specifically, step S5 involves multiple steps to achieve multi-source data fusion verification and accurate decision-making. In S51, when converting the leakage area range data retrieved by ground-penetrating radar into a geodetic coordinate system, a Gaussian projection transformation method is used, with the transformation error controlled within 0.1 meters. When converting the distance candidate set of pressure location into coordinate candidate points, five candidate location points are generated for each detection point, with the candidate point spacing set to 0.2 meters to ensure comprehensive spatial matching. During the execution of S52, the multi-source data fusion module of the early warning decision platform is called, and a weighted fusion strategy is adopted. The weight of the ground-penetrating radar inversion result is set to 0.6, and the weight of the pressure location result is set to 0.4. The two types of results are concatenated with the high-dimensional leakage feature vector to generate a comprehensive feature matrix with a dimension of 1×2048. The calculation rate of the fusion process is no less than 100 frames / second. In S53, the decision-making module performs threshold screening on the comprehensive feature matrix based on preset leakage judgment rules. The leakage feature threshold is set to 0.9, filtering out target areas exceeding the threshold. A coordinate calibration algorithm is used to determine the precise leak location coordinates, with coordinate accuracy controlled within 0.3 meters. During S54 implementation, based on the change range of media parameters, pressure loss rate, and anomaly degree of feature vector corresponding to the leak location, the leakage degree is divided into 10 levels. The level division interval is set according to the pipeline safe operation threshold. Complete leakage degree feature parameters are output through quantitative calculation. This step-by-step design, through spatial matching, data fusion, decision screening, and parameter calculation, achieves deep integration and verification of multi-source data, ensuring the accuracy and reliability of leak location and degree determination.

[0044] like Figure 2As shown, an intelligent detection and early warning location method for oil and gas pipeline leaks is implemented through different units, including: a multi-dimensional signal collaborative acquisition unit, a spectral spatial attention feature extraction unit, a ground-penetrating radar signal inversion calculation unit, a pressure curve distance location analysis unit, a multi-source data fusion decision unit, and a detection and early warning information transmission unit. The multi-dimensional signal collaborative acquisition unit and the spectral spatial attention feature extraction unit are connected through a high-speed data transmission interface to transmit the acquired multi-dimensional raw signal dataset to the feature extraction unit in real time. The spectral spatial attention feature extraction unit establishes bidirectional data interaction channels with the ground-penetrating radar signal inversion calculation unit and the pressure curve distance location analysis unit, respectively, to generate high-dimensional leak characteristics. The eigenvectors are simultaneously distributed to two computing units; the ground-penetrating radar signal inversion computing unit and the pressure curve distance positioning analysis unit are both connected to the multi-source data fusion decision unit via a data bus, transmitting their respective calculation results to the fusion decision unit for data fusion processing; the multi-source data fusion decision unit is connected to the detection and early warning information transmission unit via a communication protocol interface, transmitting the final leak location coordinates, leak degree characteristic parameters, and early warning level information to the transmission unit; the detection and early warning information transmission unit establishes a connection with the terminal monitoring system via a wireless communication network to push and provide feedback on detection and early warning information in real time. All units work collaboratively through a distributed control module, forming a complete technical link from signal acquisition to early warning output.

[0045] like Figure 3The diagram shows a real-world operation of the system according to the present invention, representing a visual representation of the invention's transformation from theoretical solution to engineering application. This on-site control terminal, serving as the front-end interactive carrier of the intelligent early warning and decision-making platform for pipeline leaks, integrates four core functions: real-time leak warning, precise location marking, multi-dimensional signal monitoring, and parameter configuration and control. It forms a deep hardware and software synergy with the multi-dimensional signal collaborative acquisition unit, multi-source data fusion decision-making unit, and detection and early warning information transmission unit described in this invention. The pop-up warning window indicating a detected pipeline leak is a direct display of the calculation results of the early warning level determination expression. The leak location coordinates marked in the pop-up window (X=125m, Y=30m) are the precise leak location output after multi-source data fusion verification in step S5 of this invention, matching the positioning accuracy index within 0.3 meters. The parameter adjustment area of ​​the control interface allows configuration of parameters such as sampling frequency, signal threshold, and detection sensitivity, perfectly matching the core parameter settings such as the 1000 Hz sampling frequency of the distributed sensing unit in S1 and the 0.05-0.2 MPa pressure change threshold in S4. The signal fluctuation graph and pipeline length scale at the bottom of the interface provide real-time rendering of multi-dimensional signal data such as vibration waves, pressure, and temperature along the pipeline, representing a visual output of data collected by the multi-dimensional signal collaborative acquisition unit. Simultaneously, the field control console, via a 5G and LoRa dual-mode communication module, links with the sensing units along the pipeline, corresponding to the wired + wireless transmission method of S6. This enables real-time data transmission and instant push of early warning information, fully replicating the entire process of signal acquisition, feature mining, inversion and localization, fusion decision-making, and early warning transmission from S1 to S6. It intuitively demonstrates the method's real-time detection and rapid early warning capabilities in complex field conditions such as on land and in the field.

[0046] like Figure 4The diagram shows the background code interface of this method. Using Python as the core programming environment, it transforms the mathematical models, algorithm flows, and technical formulas of the invention into executable software code, serving as the core software supporting the automated operation of the entire intelligent detection and early warning system. The left side of the interface is the core algorithm code editing area. The code logic strictly corresponds to the key technical modules in this invention: the feature extraction code for the spectral spatial attention detection network, demonstrating the operational logic of 3×3 convolution, spatial attention, and batch normalization in the formulas of this invention; the ground-penetrating radar signal inversion code, realizing the integral solution of the electromagnetic parameters and the simulation of signal attenuation; the pressure curve distance positioning code, completing the spatiotemporal correlation and distance mapping calculation of the pressure wave in the formulas; and the multi-source data fusion code, performing the weighted summation and 1×1 convolution feature concatenation operations in the formulas. The variable naming, function calls, and iteration logic in the code are highly consistent with the parameter definitions and algorithm steps in the invention. For example, the gradient descent algorithm implements the iterative optimization of the ground-penetrating radar inversion in S3, and the spatiotemporal matching code completes the time synchronization and spatial correlation of the pressure signal in S4. The data panels on the right and bottom of the interface display the real-time output of the code execution: the data pressure curve panel presents the pressure time series characteristics after S4 processing, intuitively displaying key parameters such as pressure peak and rising edge slope; the detection algorithm parameter panel displays configurations such as sampling frequency, iteration step size, and distance correction coefficient, which perfectly match the 0.1-meter grid step size of S3 and the 0.8-1.1 correction coefficient of S4; the leak point location coordinate panel outputs the precise location after fusion decision in real time, which is the software presentation of the S5 multi-source data fusion verification results. This code interface realizes full-process automation of signal processing, model calculation, and result visualization, transforming the technology of this invention into a deployable software system, ensuring efficient operation of multi-module collaborative work, weak leak signal identification, and precise location calculation, and is the core software carrier for transforming this method from a technical solution into an engineering application.

[0047] This invention integrates different scalar and vector parameters for unified calculation, establishing a standardized parameter mapping system and its logical association with physical meaning suitable for oil and gas pipeline leak detection scenarios. First, dimensionality normalization eliminates differences in parameter types; for example, vector-type pressure wave propagation direction vectors and scalar-type dielectric constants are converted into dimensionless values ​​in the [0,1] interval, ensuring comparability of different parameter types on a numerical scale. Second, a parameter association framework is constructed based on the physical propagation laws of pipeline leaks. For instance, in the ground-penetrating radar signal inversion formula, the vector form of the radar wave propagation path is combined with the scalar signal attenuation coefficient. Essentially, this clarifies the coupling relationship between the two in the abnormal response of the medium through electromagnetic propagation theory. The propagation path vector reflects spatial distribution characteristics, while the attenuation coefficient scalar characterizes the degree of energy loss; both jointly serve the abnormal response parameters of the medium. The algorithm performs inverse calculations of data; furthermore, it assigns reasonable contribution values ​​to different types of parameters through a weighting mechanism. For example, in the multi-source data fusion formula, the high-dimensional leakage feature vector of the vector class and the distance localization result of the scalar class are weighted according to their respective importance in leakage determination, so as to achieve complementarity between the dimensional information of vector features and the quantitative information of scalar data; finally, it utilizes the mathematical adaptability of the algorithm model, such as attention mechanism and convolution operation, to incorporate the standardized scalar and vector parameters into a unified calculation framework, which not only retains the direction and distribution information of vector parameters, but also gives full play to the quantitative representation role of scalar parameters, ultimately realizing the collaborative calculation of multiple types of parameters and accurately supporting the needs of leakage detection and localization.

[0048] The intelligent detection and early warning location method for oil and gas pipeline leaks integrates vibration wave, pressure change, and temperature field signals through multi-dimensional signal collaborative acquisition. Combined with a spectral spatial attention detection network, it achieves joint feature mining of spatial and spectral dimensions, breaking the limitations of traditional single-signal detection and solving the problem of insufficient synergy between feature extraction and signal analysis. This significantly improves the ability to identify weak leak signals in complex environments and avoids misjudgments caused by interference signals. The method utilizes a ground-penetrating radar signal inversion model to analyze changes in the electromagnetic parameters of the medium surrounding the pipeline, and combines a pressure curve distance positioning algorithm to establish a mapping relationship between pressure propagation and distance. Finally, a pipeline leak intelligent early warning decision platform completes multi-source data fusion verification, compensating for the deficiencies in positioning accuracy and data fusion efficiency of existing technologies, and achieving precise location of the leak and accurate determination of the leak's extent.

[0049] This method adopts a modular and collaborative design throughout the entire process, forming a closed-loop link from signal acquisition, feature extraction, inversion calculation, distance positioning to fusion decision-making and information transmission. The technical parameters of each link are highly compatible, ensuring the continuity and stability of detection and early warning. The organic combination of spectral spatial attention detection network, ground penetrating radar signal inversion model and pressure curve distance positioning algorithm realizes in-depth mining of multi-dimensional features and efficient integration of multi-source data, which not only improves the comprehensiveness of leak detection, but also enhances the reliability of positioning results. The integration of intelligent early warning decision platform for pipeline leaks enables rapid transmission and synchronous feedback of detection results and early warning information, meeting the core requirements of pipeline operation and maintenance for real-time performance and accuracy under complex operating conditions, and providing comprehensive technical support for the safe operation of oil and gas pipelines.

[0050] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent detection, early warning, and location of leaks in oil and gas pipelines, characterized in that, Includes the following steps: S1, through distributed sensing units, collect vibration wave signals, medium pressure change signals and pipeline surface temperature field signals along the oil and gas pipeline to construct a multi-dimensional raw signal dataset; S2, input the multi-dimensional original signal dataset into the spectral spatial attention detection network, and perform joint feature mining of the spatial and spectral dimensions of the signal through the feature extraction layer of the network to generate a high-dimensional leakage feature vector; S3, call the ground penetrating radar signal inversion model to perform inversion calculation on the abnormal response characteristics of the medium in the high-dimensional leakage feature vector, analyze the electromagnetic parameter variation law of the medium around the pipeline, and locate the preliminary range of the potential leakage area. S4. The pressure curve distance positioning algorithm is used to perform spatiotemporal correlation analysis on the characteristic parameters corresponding to the medium pressure change signal, and to establish the pressure propagation model and distance mapping relationship. S5 inputs the ground-penetrating radar signal inversion results and pressure curve distance positioning results into the pipeline leakage intelligent early warning decision platform, performs multi-source data fusion verification, and outputs accurate leakage location coordinates and leakage degree characteristic parameters. S6, through the signal transmission module of the intelligent early warning decision platform for pipeline leakage, synchronizes the leakage location coordinates, leakage degree characteristic parameters and early warning level information to the terminal monitoring system, forming a closed-loop detection and early warning process.

2. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, The feature extraction expression of the spectral spatial attention detection network is: ,in, To leak high-dimensional feature vectors, It is the Sigmoid activation function. These are the spectral dimension weighting coefficients. for Convolution operation, The original multi-dimensional signal dataset, For element-wise multiplication, The spatial dimension weight coefficients, Attention ( ) is the spatial attention computation function. For spatial attention adjustment parameters, For feature extraction bias term, For batch normalization operations, This is a max pooling operation.

3. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, The inversion calculation expression of the ground-penetrating radar signal inversion model is as follows: ,in, These are characteristic parameters of the abnormal response of the medium. The length of the pipeline inspection section. The relative permittivity of the medium surrounding the pipe. The relative permeability of the medium surrounding the pipe. This is the frequency-time domain data of the raw ground-penetrating radar signal. To detect planar coordinates, The reference coordinates for the ground-penetrating radar antenna. The signal attenuation coefficient is... For radar signal frequency, For the duration of transmission.

4. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, The distance calculation expression for the pressure curve distance positioning algorithm is as follows: ,in, The distance between the leak point and the detection point. The number of pressure detection points. For the first Each testing point Changes in pressure over time. For the first The medium flow velocity at each detection point For the first The angle between the pressure wave propagation direction at each detection point and the pipeline axis. for Pipeline position at all times Pressure value, For time intervals, This is the distance correction factor. The pressure change threshold, It is a symbolic function.

5. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, The multi-source data fusion expression of the intelligent early warning decision-making platform for pipeline leakage is: ,in, To integrate decision-making outputs, As the weights of the ground-penetrating radar inversion results, Weight the pressure location results. For the feature vector weights, For the fusion adjustment coefficient, for Convolution operation, This is a feature splicing operation.

6. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, The expression for determining the warning level of the intelligent early warning decision-making platform for pipeline leakage is as follows: ,in, At the warning level, It is the L2 norm. For the weighting coefficients of the decision outcome, The weighting coefficients are the mean of the eigenvectors. Thresholds are set for the warning level. This is a floor operation.

7. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, S3 includes the following sub-steps: S31, filtering the electromagnetic response-related features in the high-dimensional leakage feature vector, extracting a subset of features related to the changes in dielectric constant and magnetic permeability of the medium surrounding the pipeline, and establishing a mapping relationship between features and medium parameters; S32, inputting the feature subset into the initialization module of the ground-penetrating radar signal inversion model, setting the spatial grid division parameters and iterative convergence conditions for the inversion calculation, and determining the initial boundary values ​​for the inversion calculation; S33, spatially discretizing the detection area according to the set grid parameters, and simulating the propagation path and signal attenuation law of radar waves under different medium parameters through the forward modeling module of the model; S34, comparing the forward modeling simulation results with the actual collected radar signal features, and iteratively optimizing and adjusting the estimated values ​​of medium parameters until the consistency between the inversion results and the actual signal meets the set requirements.

8. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, S4 includes the following sub-steps: S41, extracting pressure time series data of each detection point from the multi-dimensional original signal dataset, removing trend components from the signal, and retaining dynamic pressure change components related to leakage; S42, extracting time-domain features from the processed pressure time series data, obtaining pressure peak value, rising slope, and fluctuation frequency calibration feature parameters, and constructing a pressure feature matrix; S43, inputting the pressure feature matrix into the pressure curve distance positioning algorithm, and establishing a calculation model for pressure wave propagation speed and distance by combining pipeline material parameters and medium physical property parameters; S44, using the algorithm's spatiotemporal matching module to synchronize and spatially correlate the pressure features of different detection points, calculating the distance estimate from each detection point to the potential leakage point, and forming a distance candidate set.

9. The intelligent detection and early warning location method for oil and gas pipeline leaks according to claim 1, characterized in that, S5 includes the following steps: S51, converting the potential leak area range data obtained from ground-penetrating radar signal inversion into regional boundary parameters under a unified spatial coordinate system, and spatially matching them with the distance candidate set obtained from pressure curve distance positioning; S52, calling the multi-source data fusion module of the pipeline leak intelligent early warning decision platform, and using a weighted fusion strategy to fuse the inversion results, positioning results, and high-dimensional leak feature vectors to generate a comprehensive feature matrix; S53, analyzing the comprehensive feature matrix through the platform's decision judgment module, and filtering out target areas that meet the leak characteristics according to preset leak judgment rules to determine the precise leak location coordinates; S54, calculating the leak degree feature parameters based on the degree of change of medium parameters and pressure loss corresponding to the precise leak location coordinates to form a complete detection and positioning result.

10. The intelligent detection and early warning location method for oil and gas pipeline leaks according to any one of claims 1-9, characterized in that, This method is implemented through different units, including: a multi-dimensional signal collaborative acquisition unit, a spectral spatial attention feature extraction unit, a ground-penetrating radar signal inversion calculation unit, a pressure curve distance positioning analysis unit, a multi-source data fusion decision unit, and a detection and early warning information transmission unit; The multi-dimensional signal collaborative acquisition unit and the spectral spatial attention feature extraction unit are connected via a high-speed data transmission interface to transmit the acquired multi-dimensional raw signal dataset to the feature extraction unit in real time. The spectral spatial attention feature extraction unit establishes bidirectional data interaction channels with the ground-penetrating radar signal inversion calculation unit and the pressure curve distance positioning analysis unit, respectively, and synchronously distributes the generated high-dimensional leakage feature vector to the two calculation units. The ground-penetrating radar signal inversion calculation unit and the pressure curve distance positioning analysis unit are both connected to the multi-source data fusion decision unit via a data bus, and transmit their respective calculation results to the fusion decision unit for data fusion processing. The multi-source data fusion decision unit is connected to the detection and early warning information transmission unit via a communication protocol interface, and transmits the final leakage location coordinates, leakage degree characteristic parameters, and early warning level information to the transmission unit. The detection and early warning information transmission unit establishes a connection with the terminal monitoring system via a wireless communication network to push and provide feedback on detection and early warning information in real time. All units work collaboratively through a distributed control module to form a complete technical link from signal acquisition to early warning output.

Images