Buried pipeline leakage signal collaborative classification system based on twin neural network

Abstract

The invention discloses a buried pipeline leakage signal collaborative classification system based on a twin neural network. The system comprises a data acquisition module, a signal preprocessing module, a twin neural network classification module, an intelligent decision module and a man-machine interaction module. The twin neural network comprises two sub-networks sharing weights, and the two sub-networks respectively process multi-modal fusion signals of the same pipe section at different moments, so that intelligent detection of pipeline state changes is realized. According to the method, the leakage detection precision can be effectively improved, the false alarm rate is reduced, accurate classification of different leakage types is realized, and reliable guarantee is provided for safe operation of buried pipelines. According to the system, the DTS is matched with the DAS technology, and intelligent identification and classification of pipeline leakage signals are realized by utilizing a laying mode that an optical cable is attached to a buried pipeline in an accompanying manner.

Description

Technical Field

[0001] This invention relates to the field of pipeline safety monitoring technology, and in particular to a collaborative classification system for buried pipeline leakage signals based on twin neural networks. Background Technology

[0002] Buried pipelines, as crucial infrastructure, are widely used in oil, natural gas, water supply, and heating. Pipeline leaks not only cause severe resource losses and environmental pollution but can also lead to major personal injury accidents. With accelerating urbanization and aging infrastructure, the need for pipeline safety monitoring is becoming increasingly urgent.

[0003] Currently, pipeline monitoring technologies mainly include traditional monitoring technologies and emerging sensing technologies. Traditional pipeline monitoring methods include pressure monitoring, flow balance, and acoustic detection, each with its own advantages and disadvantages: pressure monitoring is less sensitive to small leaks; flow balance is easily affected by flow fluctuations; and acoustic detection is easily affected by environmental noise.

[0004] In recent years, fiber optic sensing technology has been widely used in pipeline monitoring. Patent CN201821350765 discloses a water pipe leak alarm system based on fiber optic sensing, employing distributed fiber optic sensing technology to achieve continuous monitoring of long-distance pipelines. Patent CN202011317202 discloses an experimental device for improving the alarm accuracy of a fiber optic Raman temperature detection system for water pipe leaks, enhancing detection accuracy through optimization of fiber optic sensing technology. Patent CN202110340370 discloses a sensor and monitoring method suitable for long-distance subsea pipeline leak monitoring, using a vibration-sensing fiber optic interferometer to achieve high-precision detection of pipeline leaks. Patent CN202120238354 discloses a fiber optic grating oil well high-pressure water pipe leak monitoring system, utilizing a fiber optic grating temperature sensor for leak monitoring.

[0005] Distributed fiber optic sensing technology, as an emerging monitoring method, has shown great potential in the field of pipeline monitoring. This technology boasts advantages such as long sensing distance, passive operation, and strong anti-interference capabilities, and mainly includes distributed temperature sensing (DTS) and distributed acoustic sensing (DAS) technologies. DTS technology, based on the Raman scattering principle, detects pipeline leaks by monitoring temperature changes; DAS technology, based on the Rayleigh scattering effect, enables long-distance vibration monitoring.

[0006] In the field of intelligent signal processing, traditional methods mainly rely on threshold judgment and simple pattern recognition algorithms, lacking the ability to learn deep learning techniques for complex signal patterns. In recent years, deep learning technology has made breakthrough progress in the field of pattern recognition, especially Siamese neural networks, which have performed exceptionally well in similarity learning and classification tasks.

[0007] The existing technology has the following problems: 1. The problem of high false alarm rate of single sensing technology: Existing fiber optic temperature sensing technology is easily affected by ambient temperature fluctuations, and acoustic detection technology is easily affected by mechanical vibration interference. A single sensor is difficult to accurately distinguish between real leakage signals and environmental interference signals. 2. The problem of difficulty in multi-source information fusion: Existing systems lack effective methods for multi-source information fusion, and mostly rely on simple logical judgments, which cannot make full use of multi-dimensional information such as temperature and vibration for collaborative analysis; 3. Low accuracy in leak type identification: Existing technologies cannot accurately classify different leak types such as pinhole leaks, crack leaks, and joint leaks, and lack intelligent signal analysis capabilities; 4. Poor adaptability to complex environments: In urban environments with many sources of interference such as third-party construction and traffic vibrations, the existing system is not capable of identifying interference signals, and the false alarm rate is generally above 15%.

[0008] Based on this, this application proposes a collaborative classification system for buried pipeline leakage signals based on twin neural networks. Summary of the Invention

[0009] This invention provides a collaborative classification system for buried pipeline leakage signals based on twin neural networks. This system uses distributed temperature sensing (DTS) combined with distributed acoustic sensing (DAS) technology and utilizes the fiber optic cable deployment method that closely follows the buried pipeline to achieve intelligent identification and classification of pipeline leakage signals.

[0010] According to one aspect of this disclosure, a collaborative classification system for buried pipeline leakage signals based on Siamese neural networks is provided, the system comprising: Data acquisition module: used to acquire temperature signals, as well as vibration and acoustic signals; Signal preprocessing module: used to filter, extract features and standardize the acquired multi-source signals, form a unified feature representation through multimodal fusion, and perform data segmentation according to time windows; The Siamese neural network classification module consists of two sub-networks with shared weights. Each sub-network includes a feature extraction layer and a similarity judgment layer. The feature extraction layer extracts multimodal fusion feature vectors through a convolutional neural network. The similarity judgment layer outputs a similarity metric and anomaly confidence of signals at two time points through a fully connected layer and a sigmoid activation function. The two are complementary, and the similarity metric is used as an auxiliary judgment indicator to evaluate the stability of the pipeline state. Intelligent Decision Module: Used for multi-level threshold judgment: setting multi-level alarm thresholds including temperature threshold, vibration threshold, and fusion confidence threshold; Spatiotemporal correlation analysis: combining time series analysis and spatial location information to verify the spatiotemporal correlation of leakage events; False alarm suppression algorithm: establishing a false alarm filtering mechanism based on historical data statistics and environmental factor analysis; This intelligent decision module determines the type of leakage, including: large leakage, small leakage, vibration leakage, and normal state; Human-computer interaction module: used for system monitoring and parameter configuration.

[0011] In one possible implementation, during data acquisition, the distributed temperature sensing (DTS) subsystem collects temperature data via optical fiber laid along the pipeline, and the distributed acoustic sensing (DAS) subsystem collects vibration and acoustic signals via the optical fiber.

[0012] In one possible implementation, the twin neural network includes two sub-networks with shared weights, which process the normal operation reference signal and the current monitoring signal of the same pipeline segment respectively, and realize intelligent detection of pipeline status changes through similarity comparison; The specific structure of a twin neural network includes: Input layer: Receives reference signal and current monitoring signal for the same pipe segment respectively. Reference signal: temperature and vibration signal when the pipeline is operating normally at time t1. Current monitoring signal: temperature and vibration signal at the current time t2. Convolutional layer: One-dimensional convolutional kernels are used to extract local features of the signal. The kernel size is 3×1 and the stride is 1. Pooling layer: Max pooling is used to reduce data dimensionality while retaining key information. The pooling window size is 2×1. Fully connected layer: Implements high-level abstraction and non-linear mapping of features, containing 128 neurons; Similarity calculation layer: calculates the Euclidean distance and cosine similarity between the feature vectors of the reference signal and the current monitoring signal; Output layer: Outputs similarity metric and anomaly confidence. The higher the similarity metric, the more similar the signals are and the more stable the pipeline is. The higher the anomaly confidence, the more confident the neural network is in detecting an anomaly and triggering an alarm.

[0013] In one possible implementation, the signal preprocessing module includes a multimodal fusion unit and a time window segmentation unit. The multimodal fusion unit includes a noise filtering unit, a feature extraction unit, and a deep fusion processing unit. The multimodal fusion unit is used to deeply fuse the DTS temperature signal and the DAS vibration signal to form a unified multimodal feature representation. The time window segmentation unit includes a data normalization unit, a time window segmentation subunit, and a paired input data generation unit. The time window segmentation unit is used to segment the continuous signal according to a time window to provide paired input data for the Siamese neural network.

[0014] In one possible implementation, the human-machine interaction module is used for: real-time monitoring interface: displaying pipeline status, temperature distribution, vibration signals and alarm information using a graphical interface; historical data query: providing data playback, trend analysis and event query functions; system configuration management: supporting parameter setting, model update and system maintenance functions.

[0015] In one possible implementation, the optical cable is laid close to the outer wall of the pipeline and fixed every 2m using a special stainless steel clamp to ensure good heat conduction and vibration transmission with the pipeline. The outer sheath of the optical cable is made of corrosion-resistant material.

[0016] Compared with the prior art, the beneficial effects of the present invention are: Significantly improved detection accuracy: Through the collaborative work of DTS and DAS dual sensing technologies, complementary verification of temperature and vibration signals is achieved, with a detection accuracy of over 95%, which is more than 20% higher than that of single sensing technology. Significantly reduced false alarm rate: Siamese neural networks can effectively learn the characteristic differences between real leakage signals and environmental interference, reducing the false alarm rate from more than 15% in existing technologies to less than 5%, thereby improving system reliability; Precise Leakage Classification: It can accurately identify five different types of leaks, including pinhole leaks, crack leaks, joint leaks, pipe ruptures, and valve leaks, with a classification accuracy rate of over 95%, providing precise guidance for maintenance decisions. Fast response speed: The system response time is less than 30 seconds, which is more than 50% faster than traditional methods. It can detect and locate leaks in a timely manner and reduce losses. Strong environmental adaptability: Through continuous optimization of deep learning algorithms and multi-level threshold judgment, the system can adapt to different pipeline materials, burial depths and environmental conditions, and maintain stable performance in complex urban environments. Reduced construction costs: Using a single optical cable to achieve dual monitoring functions of temperature and vibration reduces construction costs by more than 30% compared to traditional multi-sensor solutions, while also reducing maintenance workload; Long monitoring distance: A single system can achieve continuous monitoring of more than 40km, which greatly improves the monitoring coverage compared with traditional point sensors; High system stability: The fiber optic sensor is passive and resistant to electromagnetic interference, and can still work stably in harsh environments, with a system availability of over 99.5%.

[0017] The buried pipeline leakage signal collaborative classification system based on twin neural networks in this disclosure can integrate multiple sensing technologies and have intelligent signal analysis capabilities to improve detection accuracy, reduce false alarm rate, and achieve accurate identification and classification of different leakage types. Attached Figure Description

[0018] Figure 1 This is a diagram illustrating the overall architecture of a collaborative classification system for buried pipeline leakage signals based on twin neural networks. Figure 2 This is a schematic diagram of a twin neural network structure; Figure 3 This is a schematic diagram of optical cable deployment. Figure 4 This is a flowchart of the signal processing procedure. Figure 5 A schematic diagram of a decision tree for leak classification; Figure 6 This is a schematic diagram of the system interface. Detailed Implementation

[0019] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0020] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0021] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0022] A collaborative classification system for buried pipeline leakage signals based on twin neural networks, such as Figure 1 As shown, the system includes: a data acquisition module, a signal preprocessing module, a Siamese neural network classification module, an intelligent decision-making module, and a human-computer interaction module; among which, Data acquisition module: Distributed temperature sensing (DTS) subsystem: Based on the principle of Raman scattering, it uses optical fiber to be laid along the pipeline to monitor temperature distribution changes in real time, with a temperature accuracy of ±0.1℃ and a spatial resolution of 1m; Distributed Acoustic Sensing (DAS) Subsystem: Based on the Rayleigh scattering effect, it uses the same optical cable to monitor vibration and acoustic signals, with a vibration frequency range of 1-20000Hz and a spatial resolution of 5m; The signal preprocessing module includes a multimodal fusion unit and a time window segmentation unit. The multimodal fusion unit includes a noise filtering unit, a feature extraction unit, and a data normalization unit, which are used to filter, extract features, and normalize the acquired multi-source signals, forming a unified feature representation through multimodal fusion. The time window segmentation unit includes a data normalization unit, a time window segmentation subunit, and a paired input data generation unit. The time window segmentation unit is used to segment continuous signals according to time windows (data segmentation processing according to time windows) to provide paired input data for the Siamese neural network.

[0023] Multimodal fusion unit: Deeply fuses DTS temperature signal and DAS vibration signal to form a unified multimodal feature representation; Time window segmentation unit: The continuous signal is segmented into time windows to provide paired input data for the twin neural network.

[0024] Siamese neural network classification module: Feature extraction layer (neural network architecture): Extracts multimodal fusion feature vectors through convolutional neural networks; Similarity judgment layer: Through a fully connected layer and a sigmoid activation function, it outputs the similarity measure and anomaly confidence of signals at two time points.

[0025] Adaptive Multi-Data Fusion and Network Parameter Description: Modal data: Each time step includes DTS temperature data (L×T dimension) and DAS vibration data (L / 5×T×F dimension). Feature fusion: DTS spatiotemporal resampling + DAS frequency domain feature extraction, fused into a feature vector of unified dimension D; Twin inputs: The reference signal (t1) and the current monitoring signal (t2) are respectively input into two symmetrical branches of the network; Similarity learning: The network is trained to identify signal-like differences between normal states and leakage anomalies through contrastive learning; Shared subnetwork: Dense (D / 2) Dense (D / 4) Dense(D / 8), a weighted similarity measure: Euclidean distance is used to calculate the difference between D / 8 dimensional feature vectors; Determine the network type: Dense (D / 16) Dense(1) outputs the similarity score; Training strategy: Contrastive learning, maximizing the similarity of positive samples and minimizing the similarity of negative samples.

[0026] Activation functions: ReLU (feature layer), Tanh (encoding layer), Sigmoid (output layer).

[0027] Intelligent decision-making module: Multi-level threshold judgment: Set multiple alarm thresholds such as temperature threshold, vibration threshold, and fusion confidence threshold; Spatiotemporal correlation analysis: Combining time series analysis and spatial location information, the spatiotemporal correlation of the leakage event is verified; False alarm suppression algorithm: A false alarm filtering mechanism is established based on historical data statistics and environmental factor analysis.

[0028] Human-computer interaction module: Real-time monitoring interface: A graphical interface is used to display pipeline status, temperature distribution, vibration signals and alarm information; Historical data query: Provides data playback, trend analysis, and event query functions; System configuration management: Supports parameter setting, model update and system maintenance functions.

[0029] The specific structure of the twin neural network includes: Input layer: Receives reference signals (at time t1: DTS temperature signal and DAS vibration signal during normal pipeline operation) and current monitoring signals (at time t2: DTS temperature signal and DAS vibration signal at the current time) from the same pipeline segment respectively. Convolutional layer: One-dimensional convolutional kernels are used to extract local features of the signal. The kernel size is 3×1 and the stride is 1. Pooling layer: Max pooling is used to reduce data dimensionality while retaining key information. The pooling window size is 2×1. Fully connected layer: Implements high-level abstraction and non-linear mapping of features, containing 128 neurons; Similarity calculation layer: calculates the Euclidean distance and cosine similarity between the feature vectors of the reference signal and the current signal; Output layer: Outputs similarity metric (the higher the value, the more similar the signals and the more stable the pipeline state) and anomaly confidence (the higher the value, the more confident the neural network is in detecting an anomaly and triggering an alarm).

[0030] The optical cable is laid as follows: the optical cable is laid close to the outer wall of the pipeline, and fixed every 2 meters using special stainless steel clamps to ensure good heat conduction and vibration transmission with the pipeline. The outer sheath of the optical cable is made of corrosion-resistant material. Laying parameters: Optical cable structure: one cable contains two fiber cores; Fiber core structure: one multimode and one single-mode fiber. Installation method: close to the pipe wall; monitoring distance: 10km; spatial resolution: 1m; temperature accuracy: ±0.1℃; vibration frequency range: 1-20000Hz; response time: less than or equal to 5 seconds.

[0031] The system's signal processing flow (processing parameters: window length 1024, overlap rate 50%, confidence threshold 0.8, filtering frequency 0.1-500Hz, response time: ≤5s, detection accuracy 95%) is as follows: Figure 4 As shown: Data acquisition (reference signal, current signal); Data preprocessing (multimodal signal fusion, establishment of a reference signal library and pairing with the current signal); Feature extraction (reference signal-current signal difference features, multimodal fusion feature vector); Siamese neural network classification: compare the baseline signal with the current signal, fuse multimodal features, and output similarity measurement; When the system detects an anomaly with a confidence level greater than the threshold, it directly enters the signal analysis phase. Through comparison with the baseline signal and the current signal, multimodal feature change analysis, and similarity measurement evaluation (high values ​​indicate stability), it automatically identifies the anomaly type and performs leakage type identification, location positioning, and severity assessment. When the confidence level is less than or equal to the threshold, the system initiates a spatiotemporal correlation analysis mechanism. This involves secondary verification through comparison of signals from adjacent points, time series consistency checks, and false alarm suppression algorithms. Verification is successful when the signal change pattern of adjacent points does not conform to leakage propagation characteristics, the duration of the anomaly signal is too short (<30 seconds) or intermittent, or the signal change pattern matches known interference false alarm characteristics.

[0032] When the spatiotemporal correlation analysis verification passes, the system is determined to be in a normal state and continues to monitor; when the verification fails, the system enters the signal analysis (comparison of the reference signal and the current signal, changes in multimodal features, and similarity measurement (high value = stable)) stage.

[0033] The system, through a signal analysis phase, outputs temperature and vibration information analysis results to automatically identify anomaly types. For temperature anomalies, the system first analyzes the magnitude and range of temperature change. When the temperature change exceeds 5°C and the range of influence is greater than 50 meters, it is determined to be a major leak. When the temperature change is within the range of 2-5°C or the range of influence is less than 50 meters, it is determined to be a minor leak and enters the confidence level subdivision process.

[0034] In the confidence level segmentation of small leaks, an alarm is triggered directly when the confidence level is greater than 0.9; when the confidence level is between 0.7 and 0.9, secondary verification is required, and the system will re-enter the spatiotemporal correlation analysis process; when the confidence level is less than or equal to 0.7, it indicates that the similarity is high, and the system continues to monitor.

[0035] For abnormal vibration, the system makes a judgment based on the pipe type and vibration intensity. When the vibration intensity of steel pipe exceeds 50 dB or plastic pipe exceeds 40 dB, and the acceleration is greater than 0.5g and the frequency exceeds 100Hz, it is judged as a vibration leak and immediately addressed. When the vibration intensity of steel pipe is less than or equal to 50 dB or plastic pipe is less than or equal to 40 dB, and the acceleration is less than 0.1g and the frequency is less than 50Hz, it is judged as normal vibration and monitoring continues.

[0036] The entire system ensures the accuracy and reliability of leak detection through multi-level confidence assessment, signal analysis, and spatiotemporal correlation verification, effectively reducing the false alarm rate and improving the system's practicality.

[0037] Example 1: As Figure 1 As shown, the buried pipeline leakage signal collaborative classification system based on twin neural networks of the present invention includes a data acquisition module, a signal preprocessing module, a twin neural network classification module, an intelligent decision-making module, and a human-computer interaction module.

[0038] The specific implementation of the data acquisition module: The DTS subsystem employs the Raman scattering principle, with a laser wavelength of 1550nm, a temperature accuracy of ±0.1℃, and a spatial resolution of 1m. The DAS subsystem adopts the principle of coherent Rayleigh scattering, with a vibration frequency range of 1-20000Hz and a spatial resolution of 5m. The specific steps of signal preprocessing are as follows: Multimodal fusion: The DTS temperature signal and DAS vibration signal are registered and fused according to spatial location to form a unified multimodal feature representation; Time window segmentation: The continuous signal is divided into time windows according to a fixed time interval (e.g., 30 seconds), a normal operation reference signal library is established, and the current monitoring signal is paired with the reference signal for input.

[0039] The training process of a twin neural network: Construct a training dataset containing normal signal pairs and abnormal signal pairs. The normal pairs consist of the baseline signal and stable monitoring signal during normal pipeline operation, while the abnormal pairs consist of the normal baseline signal and the changes in the monitoring signal during leakage. The contrastive loss function is: L = (1-Y)×D² + Y×max(0, margin-D)², where Y=0 represents similar pairs and Y=1 represents dissimilar pairs; Using the Adam optimizer, the learning rate was set to 0.001; The training rounds are 200, and the batch size is 32.

[0040] Example 2: Adaptability optimization for different pipeline materials and environmental conditions: Steel pipelines: Temperature threshold set to ±2℃, vibration threshold set to 50dB; Plastic pipelines: Temperature threshold set to ±1℃, vibration threshold set to 40dB; Composite pipelines: An adaptive threshold algorithm is used to dynamically adjust based on historical data.

[0041] Example 3: System Deployment and Operation Maintenance Fiber optic cable installation: Stainless steel clamps are used to fix the cable every 2 meters to ensure tight contact with the pipeline; Equipment configuration: The main control unit adopts an industrial-grade computer and supports an operating temperature range of -40℃ to +70℃; Communication interface: Supports Ethernet and 4G / 5G wireless communication for remote monitoring; Time parameter settings: Reference signal establishment period: A stable signal for 7 consecutive days during the initial stage of system operation is used as the reference; Monitoring signal sampling period: The current signal is collected every 30 seconds; Comparison analysis frequency: The current signal is compared with the reference signal using a twin neural network every 30 seconds; Reference signal update cycle: The reference signal library is updated every 30 days to remove abnormal data; Maintenance strategy: Regularly calibrate the sensors and update the neural network model parameters.

[0042] Through the above technical solution, the present invention achieves high-precision detection and classification of leaks in buried pipelines, providing reliable technical support for the safe operation of pipelines.

[0043] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A collaborative classification system for buried pipeline leakage signals based on Siamese neural networks, characterized in that, The system includes: Data acquisition module: used to acquire temperature signals, as well as vibration and acoustic signals; Signal preprocessing module: used to filter, extract features and standardize the acquired multi-source signals, form a unified feature representation through multimodal fusion, and perform data segmentation according to time windows; The Siamese neural network classification module consists of two sub-networks with shared weights. Each sub-network includes a feature extraction layer and a similarity judgment layer. The feature extraction layer extracts multimodal fusion feature vectors through a convolutional neural network. The similarity judgment layer outputs a similarity metric and anomaly confidence of signals at two time points through a fully connected layer and a sigmoid activation function. The two are complementary, and the similarity metric is used as an auxiliary judgment indicator to evaluate the stability of the pipeline state. Intelligent Decision Module: Used for multi-level threshold judgment: setting multi-level alarm thresholds including temperature threshold, vibration threshold, and fusion confidence threshold; Spatiotemporal correlation analysis: combining time series analysis and spatial location information to verify the spatiotemporal correlation of leakage events; False alarm suppression algorithm: establishing a false alarm filtering mechanism based on historical data statistics and environmental factor analysis; This intelligent decision module determines the type of leakage, including: large leakage, small leakage, vibration leakage, and normal state; Human-computer interaction module: used for system monitoring and parameter configuration.

2. The buried pipeline leakage signal collaborative classification system based on Siamese neural network according to claim 1, characterized in that, During data acquisition, the distributed temperature sensing (DTS) subsystem collects temperature data through optical cables laid along the pipeline, while the distributed acoustic sensing (DAS) subsystem collects vibration and acoustic signals through the same optical cables.

3. The buried pipeline leakage signal collaborative classification system based on Siamese neural network according to claim 1, characterized in that, The twin neural network includes two sub-networks with shared weights, which process the normal operation reference signal and the current monitoring signal of the same pipeline segment respectively, and realize intelligent detection of pipeline status changes through similarity comparison. The specific structure of a twin neural network includes: Input layer: Receives reference signal and current monitoring signal for the same pipe segment respectively. Reference signal: temperature and vibration signal when the pipeline is operating normally at time t1. Current monitoring signal: temperature and vibration signal at the current time t2. Convolutional layer: One-dimensional convolutional kernels are used to extract local features of the signal. The kernel size is 3×1 and the stride is 1. Pooling layer: Max pooling is used to reduce data dimensionality while retaining key information. The pooling window size is 2×1. Fully connected layer: Implements high-level abstraction and non-linear mapping of features, containing 128 neurons; Similarity calculation layer: calculates the Euclidean distance and cosine similarity between the feature vectors of the reference signal and the current monitoring signal; Output layer: Outputs similarity metric and anomaly confidence. The higher the similarity metric, the more similar the signals are and the more stable the pipeline is. The higher the anomaly confidence, the more confident the neural network is in detecting an anomaly and triggering an alarm.

4. The buried pipeline leakage signal collaborative classification system based on Siamese neural network according to claim 1, characterized in that, The signal preprocessing module includes a multimodal fusion unit and a time window segmentation unit. The multimodal fusion unit includes a noise filtering unit, a feature extraction unit, and a deep fusion processing unit. The multimodal fusion unit is used to deeply fuse the DTS temperature signal and the DAS vibration signal to form a unified multimodal feature representation. The time window segmentation unit includes a data normalization unit, a time window segmentation subunit, and a paired input data generation unit. The time window segmentation unit is used to segment the continuous signal according to time windows to provide paired input data for the Siamese neural network.

5. The buried pipeline leakage signal collaborative classification system based on Siamese neural network according to claim 1, characterized in that, The human-machine interaction module is used for: real-time monitoring interface: displaying pipeline status, temperature distribution, vibration signals and alarm information using a graphical interface; historical data query: providing data playback, trend analysis and event query functions; system configuration management: supporting parameter setting, model update and system maintenance functions.

6. The buried pipeline leakage signal collaborative classification system based on Siamese neural network according to claim 1, characterized in that, The optical cable is laid close to the outer wall of the pipeline and fixed every 2m with special stainless steel clamps to ensure good heat conduction and vibration transmission with the pipeline. The outer sheath of the optical cable is made of corrosion-resistant material.

Images