Pipeline Leak Detection Method and Device Based on Distributed Fiber Optic Sensing Signal and Feature Joint Input FC-ANN Network
By using an FC-ANN network model based on the joint input of distributed optical fiber sensing signals and features, the problem of insufficient accuracy in pipeline leak detection in existing technologies is solved, enabling efficient identification and location of leak points in complex environments, and improving the accuracy and reliability of detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2024-07-18
- Publication Date
- 2026-06-30
AI Technical Summary
Existing pipeline leak detection methods suffer from low detection efficiency, poor positioning accuracy, and slow response speed in urban environments. Furthermore, traditional fiber optic sensing systems struggle to effectively distinguish between interference signals and leak signals, resulting in insufficient accuracy.
A fully connected artificial neural network (FC-ANN) model based on the joint input of distributed optical fiber sensing signals and features is adopted. Through filtering preprocessing, time-domain and frequency-domain feature extraction, a joint sequence is constructed, and the model training is optimized by combining evaluation indicators to achieve efficient classification of optical fiber signals.
It improves the accuracy and reliability of pipeline leak detection, effectively identifies and locates leak points in complex environments, enhances the signal-to-noise ratio and the learning effect of neural networks, and achieves a higher leak identification accuracy than traditional methods.
Smart Images
Figure CN118940169B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of optical fiber sensing technology, specifically relating to a pipeline leak detection method and device based on a distributed optical fiber sensing signal and feature joint input FC-ANN network. Background Technology
[0002] In the process of modern urbanization, urban water pipelines, as a vital component of the city's lifeline, bear the important responsibility of supplying water to the city. Their stable and efficient operation directly affects the quality of residents' daily lives and the sustainable development of the city's economy. Water pipelines consist of multiple interconnected sections, and leaks can occur at these joints due to aging, geological subsidence, construction, and other reasons, posing a serious threat to urban safety. Therefore, timely detection and accurate location of pipeline leaks are crucial for maintaining urban safety and ensuring water supply services. Traditional pipeline leak detection methods, such as manual inspections and pressure monitoring, suffer from low detection efficiency, poor location accuracy, and slow response times, making them unsuitable for the needs of modern urban management.
[0003] With the rapid development of optoelectronic technology, Distributed Optical Fiber Sensing Systems (DOFS) have been widely used in perimeter security, pipeline leak monitoring, and other fields. Among them, the Phase-Sensitive Optical Time-Domain Reflectometer (φ-OTDR) is characterized by high sensitivity, the ability to sense vibrations caused by leaks, and distributed sensing, showing great application potential in leak detection of urban water supply pipelines. In pipeline leak scenarios, the impact of water flow or pressure changes at the leak point will cause minute vibrations in the surrounding medium. These vibrations are transmitted to the optical fiber through the soil or pipe wall, captured by the φ-OTDR system, and converted into analyzable data signals.
[0004] However, despite the excellent performance of the φ-OTDR system in data acquisition, it still faces many challenges in data analysis and processing. On the one hand, the identification of pipeline leaks is highly susceptible to various interference events and noise. Urban environments contain diverse vibration sources, such as traffic vibrations, construction noise, and geological activity. These interference signals intertwine with the leak signals, making it extremely difficult to extract effective leak information from the complex background. The presence of multiple interference events and their indistinct characteristics result in low accuracy for traditional feature-based Support Vector Machine (SVM) classifiers. On the other hand, while ordinary Artificial Neural Network (ANN) models possess powerful capabilities in signal processing and classification, automatically extracting and learning features from large amounts of data to achieve high-precision classification, exceeding the accuracy of SVM classifiers, their feature extraction process often lacks specificity when applied to pipeline leak detection. This leads to low correlation between the extracted information and the leak signal features, limiting the classification effect.
[0005] Therefore, in order to further improve the accuracy and reliability of fiber optic sensing technology in pipeline leak detection, it is urgent to develop a more accurate pipeline leak detection method. Summary of the Invention
[0006] In view of the above, the purpose of this invention is to provide a pipeline leak detection method and device based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network. This method combines the advantages of a fully connected artificial neural network (FC-ANN) model and feature extraction methods to more accurately classify interference events with fuzzy features and leak events with stable features, making it suitable for efficient and accurate identification and location of pipeline leak events.
[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solution:
[0008] In a first aspect, embodiments of the present invention provide a pipeline leak detection method based on a distributed optical fiber sensing signal and feature joint input FC-ANN network, comprising the following steps:
[0009] Acquire fiber optic signals including those related to pipeline perimeter disturbances and leaks, and perform filtering preprocessing;
[0010] Feature extraction, including time-domain and frequency-domain features, is performed on the filtered preprocessed fiber optic signal, and the filtered preprocessed fiber optic signal and the extracted features are constructed into a joint sequence;
[0011] Construct a fully connected artificial neural network model, input the joint sequence into the fully connected artificial neural network model for model training, and optimize the model training effect by combining evaluation metrics;
[0012] A pre-trained fully connected artificial neural network model is used to detect pipeline leaks in newly acquired optical fiber signals.
[0013] Preferably, the acquisition of fiber optic signals including pipeline perimeter disturbances and leakage conditions includes:
[0014] Phase-sensitive distributed optical fiber sensing systems are used to collect optical fiber signals of perimeter disturbances, slow leakage, and fast leakage at the pipeline perimeter.
[0015] Preferably, the filtering preprocessing method includes:
[0016] A moving average with a window length of N is applied to the fiber optic signal with a sampling rate of fs to estimate its DC and low-frequency components. These components are then subtracted from the fiber optic signal to remove DC and low-frequency noise from the signal.
[0017] Preferably, feature extraction, including time-domain and frequency-domain features, is performed on the filtered preprocessed fiber optic signal, including:
[0018] Time-domain features are extracted from the filtered and preprocessed fiber optic signal. The time-domain features include at least one of the following: time-domain maximum value, time-domain minimum value, average value, peak-to-peak value, absolute average value, variance, standard deviation, root mean square value, root square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, and margin factor.
[0019] After filtering and preprocessing, the fiber optic signal is subjected to Fourier transform to extract frequency domain features, which include at least one of the following: average frequency, centroid frequency, root mean square frequency, and frequency standard deviation.
[0020] Preferably, the method for combining the filtered preprocessed fiber optic signal and the extracted features includes:
[0021] The filtered and preprocessed fiber signal is resampled to obtain a signal sequence with a reduced sequence length. Then, the extracted features are added to the end of the signal sequence to obtain a joint sequence of the signal and features.
[0022] Preferably, the network structure of the fully connected artificial neural network model includes:
[0023] The system consists of a linear input layer, a first dropout layer, a linear hidden layer, a second dropout layer, a linear output layer, and a softmax layer connected in sequence. ReLU activation functions are connected between the linear input layer and the first dropout layer, as well as between the linear hidden layer and the second dropout layer. The final pipeline leakage probability result is output by the softmax layer.
[0024] Preferably, the evaluation metrics used during model training include:
[0025] Accuracy, precision, recall, F-score, and cross-entropy loss.
[0026] Secondly, in order to achieve the above-mentioned objectives, the present invention also provides a pipeline leak detection device based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network, comprising: a data acquisition and preprocessing module, a signal and feature joint construction module, a neural network model training module, and a pipeline leak detection application module;
[0027] The data acquisition and preprocessing module is used to acquire fiber optic signals, including those related to pipeline perimeter disturbances and leaks, and to perform filtering and preprocessing.
[0028] The signal and feature joint construction module is used to extract features, including time-domain features and frequency-domain features, from the filtered and preprocessed optical fiber signal, and to construct a joint sequence from the filtered and preprocessed optical fiber signal and the extracted features.
[0029] The neural network model training module is used to construct a fully connected artificial neural network model, input the joint sequence into the fully connected artificial neural network model for model training, and optimize the model training effect by combining evaluation metrics.
[0030] The pipeline leak detection application module is used to detect pipeline leaks by using a trained fully connected artificial neural network model on newly acquired optical fiber signals.
[0031] Thirdly, to achieve the above-mentioned objectives, embodiments of the present invention also provide a pipeline leak detection device based on a distributed optical fiber sensor signal and feature joint input FC-ANN network, including a memory and one or more processors. The memory is used to store a computer program, and the processor is used to implement the above-mentioned pipeline leak detection method based on a distributed optical fiber sensor signal and feature joint input FC-ANN network when the computer program is executed.
[0032] Fourthly, to achieve the above-mentioned objectives, embodiments of the present invention also provide a computer-readable storage medium storing a computer program. When the computer program is executed by a computer, the above-mentioned pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network is implemented.
[0033] Compared with the prior art, the beneficial effects of the present invention include at least the following:
[0034] (1) This invention can effectively remove DC and low-frequency components and retain the effective signal quantity as much as possible by filtering and preprocessing the signal. The characteristics of the filtered signal are more obvious, which helps to improve the signal-to-noise ratio of the positioning and the learning effect of the neural network model. It can realize the monitoring of leakage in the case of a lot of interference along the water pipeline.
[0035] (2) This invention inputs fiber optic signals and features into an FC-ANN network, utilizing the effective reflection of leakage events by the feature quantities to assist the neural network in learning more effective leakage event information. At the same time, the input of the signal assists the neural network in identifying interference events. Since the time-domain waveform and time-frequency domain features of the signal are multimodal descriptions of the event to be detected, the joint input of the signal and features can more accurately reflect the event and improve the performance of the model. Each input data and output data of the FC-ANN network are connected, which is suitable for situations where each input data needs to be fully utilized. This fully combines the advantages of features, signals and FC-ANN networks, and can achieve a higher leakage identification accuracy than traditional methods. Attached Figure Description
[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is a schematic flowchart of the pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network provided in an embodiment of the present invention.
[0038] Figure 2 This is a schematic diagram of the feature extraction and joint method provided in the embodiments of the present invention;
[0039] Figure 3 This is a schematic diagram of the FC-ANN model structure provided in an embodiment of the present invention;
[0040] Figure 4This is a schematic diagram of the structure of a pipeline leak detection device based on a distributed optical fiber sensing signal and feature joint input FC-ANN network provided in an embodiment of the present invention. Detailed Implementation
[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.
[0042] The inventive concept of this invention is as follows: Addressing the problem of insufficient accuracy in existing fiber optic sensing methods for detecting pipeline leaks, this invention provides a pipeline leak detection method and apparatus based on a distributed fiber optic sensing signal and feature-based FC-ANN network. By combining fiber optic signal data and features and training a suitable FC-ANN network model, leak detection of water pipelines along the fiber optic cable can be achieved. Compared to traditional SVM classifiers and neural network models with only signal input, this effectively improves the accuracy of leak event identification.
[0043] Figure 1 This is a schematic flowchart of a pipeline leak detection method based on a distributed optical fiber sensor signal and feature joint input FC-ANN network provided in an embodiment of the present invention. Figure 1 As shown in the embodiment, a pipeline leak detection method based on a distributed optical fiber sensing signal and feature joint input FC-ANN network is provided, including the following steps:
[0044] S1 collects fiber optic signals including pipeline perimeter disturbances and leakage conditions, and performs filtering preprocessing.
[0045] S1.1, Fiber Optic Signal Acquisition. A phase-sensitive distributed fiber optic sensing system (φ-OTDR) is used to acquire fiber optic signals of perimeter disturbances, slow leakage, and fast leakage at the pipeline perimeter.
[0046] In this embodiment, the fiber length is 5000m, a spatial point is collected every 2m, the time sampling rate is 1500Hz, and the perimeter disturbance includes but is not limited to background noise, human knocking and human shaking. Slow leakage is defined as intermittent leakage, and fast leakage is defined as multiple consecutive leakage.
[0047] S1.2, Filtering Preprocessing. The filtering preprocessing method involves performing a moving average of the fiber optic signal with a sampling rate of fs and a window length of N to estimate its DC and low-frequency components. These components are then subtracted from the fiber optic signal to remove DC and low-frequency noise from the signal. This is equivalent to performing a high-pass filter of fs / N on the signal. Compared to directly using a filter for high-pass filtering, this method can remove the specified low-frequency components to the maximum extent without affecting the high-frequency components carrying the sensing information.
[0048] In this embodiment, a window of length 100 is used to perform a moving average on the time-domain signal. The processing result serves as the overall fluctuation trend of the signal. Since it is not caused by leakage, the low-frequency noise caused by signal fluctuation is removed by subtracting the moving average result from the signal.
[0049] S2 extracts features, including time-domain and frequency-domain features, from the filtered preprocessed fiber signal and constructs a joint sequence from the filtered preprocessed fiber signal and the extracted features.
[0050] S2.1 Feature Extraction. Time-domain features include at least one of the following: time-domain maximum, time-domain minimum, average, peak-to-peak value, absolute average, variance, standard deviation, root mean square value, root mean square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, and margin factor. Extracted frequency-domain features include at least one of the following: average frequency, centroid frequency, root mean square frequency, and frequency standard deviation.
[0051] In the embodiments, such as Figure 2 As shown, the 14 time-domain features are extracted after normalizing the filtered and preprocessed fiber signal, and the 4 frequency-domain features are extracted from the result after performing a Fourier transform on the normalized signal.
[0052] S2.2, Signal and Feature Combination. The filtered and preprocessed fiber optic signal is resampled to obtain a signal sequence with a reduced sequence length. Then, the extracted features are sequentially added to the end of the signal sequence to obtain a combined sequence of signal and features.
[0053] In this embodiment, the time-domain signal at the location of leakage or other disturbance is segmented to obtain data with a duration of 1 / 3s and a data size of 1×500. Then, the signal is resampled to obtain a signal with a sampling rate of 1200Hz and a length of 400. Finally, 18 extracted features are added sequentially to the end of the signal sequence to obtain a one-dimensional joint sequence of signal + features.
[0054] S3. Construct a fully connected artificial neural network model, input the joint sequence into the fully connected artificial neural network model for model training, and optimize the model training effect by combining evaluation metrics.
[0055] S3.1, Model Building. For example... Figure 3 As shown, the network structure of the FC-ANN model includes: a linear input layer, a first dropout layer, a linear hidden layer, a second dropout layer, a linear output layer, and a softmax layer connected in sequence. ReLU activation functions are connected between the linear input layer and the first dropout layer, and between the linear hidden layer and the second dropout layer. The softmax layer outputs the final pipeline leakage probability result. The ReLU activation function introduces non-linear fitting capabilities into the model; the first and second dropout layers reduce the risk of overfitting; the linear hidden layer transforms and processes the input data to better segment it; and the softmax layer normalizes the linear output layer data to obtain the probability result.
[0056] S3.2, Model Training. In this embodiment, the following is used: Figure 3 The FC-ANN model shown is trained and tested using joint sequences as the training and testing sets. There are 50 training epochs, a learning rate of 0.001, and a batch size of 100. Accuracy, precision, recall, F-score, and cross-entropy loss are used as evaluation metrics to measure model performance. The model with the best performance on the test set is selected as the pipeline leak detection model.
[0057] S4 uses a trained fully connected artificial neural network model to detect pipeline leaks in newly acquired fiber optic signals.
[0058] In this embodiment, the model is deployed in a pipeline leak monitoring device to identify abnormal energy data, and to determine whether there is a leak and provide information on the rate of leakage based on the identification results.
[0059] In summary, a pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network can effectively remove DC and low-frequency components while retaining effective signal quantities. The filtered signal features are more pronounced, which helps improve the signal-to-noise ratio for localization and the learning effect of the neural network model. This method can monitor leaks along water pipelines where there is significant interference. Furthermore, it can utilize the effective reflection of leak events by feature quantities to assist the neural network in learning more effective leak event information. By fully combining the advantages of features, signals, and the FC-ANN network, a leak identification accuracy higher than that of traditional methods can be achieved.
[0060] Based on the same inventive concept, such as Figure 4As shown, this embodiment of the invention also provides a pipeline leak detection device 400 based on a distributed optical fiber sensing signal and feature joint input FC-ANN network, including: a data acquisition and preprocessing module 410, a signal and feature joint construction module 420, a neural network model training module 430, and a pipeline leak detection application module 440.
[0061] The data acquisition and preprocessing module 410 is used to acquire fiber optic signals, including those related to pipeline perimeter disturbances and leaks, and to perform filtering and preprocessing.
[0062] The signal and feature joint construction module 420 is used to extract features, including time-domain features and frequency-domain features, from the filtered preprocessed optical fiber signal, and to construct a joint sequence from the filtered preprocessed optical fiber signal and the extracted features.
[0063] The neural network model training module 430 is used to construct a fully connected artificial neural network model. The joint sequence is input into the fully connected artificial neural network model for model training, and the model training effect is optimized by combining evaluation metrics.
[0064] The pipeline leak detection application module 440 is used to detect pipeline leaks by using a trained fully connected artificial neural network model on newly acquired fiber optic signals.
[0065] Based on the same inventive concept, this invention also provides a pipeline leak detection device based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network, including a memory and one or more processors. The memory is used to store a computer program, and the processor is used to implement the above-mentioned pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network when the computer program is executed.
[0066] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program. When the computer program is executed by a computer, the above-described pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network is implemented.
[0067] It should be noted that the pipeline leak detection device, pipeline leak detection equipment, and computer-readable storage medium based on the distributed optical fiber sensor signal and feature joint input FC-ANN network provided in the above embodiments all belong to the same inventive concept as the pipeline leak detection method based on the distributed optical fiber sensor signal and feature joint input FC-ANN network. For details of their specific implementation process, please refer to the embodiments of the pipeline leak detection method based on the distributed optical fiber sensor signal and feature joint input FC-ANN network, which will not be repeated here.
[0068] The specific embodiments described above illustrate the technical solution and beneficial effects of the present invention in detail. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A pipeline leak detection method based on a distributed optical fiber sensing signal and feature joint input FC-ANN network, characterized in that, Includes the following steps: Acquire fiber optic signals including those related to pipeline perimeter disturbances and leaks, and perform filtering preprocessing; Feature extraction, including time-domain and frequency-domain features, is performed on the filtered preprocessed fiber optic signal, and the filtered preprocessed fiber optic signal and the extracted features are constructed into a joint sequence; One of the methods for combining the filtered preprocessed optical fiber signal and the extracted features includes: resampling the filtered preprocessed optical fiber signal to obtain a signal sequence with a reduced sequence length after resampling, and then sequentially adding the extracted features to the end of the signal sequence to obtain a joint sequence of the signal and features; A fully connected artificial neural network model is constructed. The joint sequence is input into the fully connected artificial neural network model for model training, and the model training effect is optimized by combining evaluation metrics. The network structure of the fully connected artificial neural network model includes: a linear input layer, a first dropout layer, a linear hidden layer, a second dropout layer, a linear output layer, and a softmax layer connected in sequence. The ReLU activation function is connected between the linear input layer and the first dropout layer, as well as between the linear hidden layer and the second dropout layer. The final pipeline leakage probability result is output by the softmax layer. A pre-trained fully connected artificial neural network model is used to detect pipeline leaks in newly acquired optical fiber signals.
2. The pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network according to claim 1, characterized in that, The acquisition includes fiber optic signals of pipeline perimeter disturbances and leaks, including: Phase-sensitive distributed optical fiber sensing systems are used to collect optical fiber signals of perimeter disturbances, slow leakage, and fast leakage at the pipeline perimeter.
3. The pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network according to claim 1, characterized in that, Filter preprocessing methods include: For sampling rate fs The fiber optic signal is processed with a window length of... N The moving average is used to estimate its DC and low-frequency components, and these components are then subtracted from the fiber optic signal to remove DC and low-frequency noise from the signal.
4. The pipeline leak detection method based on the joint input of distributed optical fiber sensing signals and features into an FC-ANN network according to claim 1, characterized in that, Feature extraction, including time-domain and frequency-domain features, is performed on the pre-filtered fiber optic signal, including: Time-domain features are extracted from the filtered and preprocessed fiber optic signal. The time-domain features include at least one of the following: time-domain maximum value, time-domain minimum value, average value, peak-to-peak value, absolute average value, variance, standard deviation, root mean square value, root square amplitude, kurtosis, skewness, waveform factor, peak factor, impulse factor, and margin factor. After filtering and preprocessing, the fiber optic signal is subjected to Fourier transform to extract frequency domain features, which include at least one of the following: average frequency, centroid frequency, root mean square frequency, and frequency standard deviation.
5. The pipeline leak detection method based on the joint input of distributed optical fiber sensor signals and features into an FC-ANN network according to claim 1, characterized in that, The evaluation metrics used during model training include: Accuracy, precision, recall, F-score, and cross-entropy loss.
6. A pipeline leak detection device based on a distributed optical fiber sensor signal and feature joint input FC-ANN network, implemented using the pipeline leak detection method based on a distributed optical fiber sensor signal and feature joint input FC-ANN network as described in any one of claims 1 to 5, characterized in that, include: The system includes a data acquisition and preprocessing module, a signal and feature joint construction module, a neural network model training module, and a pipeline leak detection application module. The data acquisition and preprocessing module is used to acquire fiber optic signals, including those related to pipeline perimeter disturbances and leaks, and to perform filtering and preprocessing. The signal and feature joint construction module is used to extract features, including time-domain features and frequency-domain features, from the filtered and preprocessed optical fiber signal, and to construct a joint sequence from the filtered and preprocessed optical fiber signal and the extracted features. The neural network model training module is used to construct a fully connected artificial neural network model, input the joint sequence into the fully connected artificial neural network model for model training, and optimize the model training effect by combining evaluation metrics. The pipeline leak detection application module is used to detect pipeline leaks by using a trained fully connected artificial neural network model on newly acquired optical fiber signals.
7. A pipeline leak detection device based on a distributed optical fiber sensing signal and feature joint input FC-ANN network, comprising a memory and one or more processors, wherein the memory is used to store a computer program, characterized in that, The processor is used to implement the pipeline leak detection method based on the distributed optical fiber sensing signal and feature joint input FC-ANN network as described in any one of claims 1-5 when executing the computer program.
8. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a computer, it implements the pipeline leakage detection method based on the distributed optical fiber sensing signal and feature joint input FC-ANN network as described in any one of claims 1-5.