An acoustic signal intelligent diagnosis method and system for oil and gas pipeline network digital twinning
Patent Information
- Application Number
- CN202610931458.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-26
- Publication Date
- 2026-10-09
AI Technical Summary
[0006]为了解决现有油气管道声波泄漏检测方法在微小泄漏识别能力不足以及数字孪生系统智能诊断能力有限的问题
[0053]本发明将深度学习技术与数字孪生系统相结合，引入层级多尺度特征补偿网络，实现对油气管道声波信号的智能分析与泄漏异常检测。通过层级多尺度特征建模结构，使网络能够逐层提取声波信号在不同时间尺度下的特征信息，有效解决了传统CNN感受野有限、难以同时兼顾局部瞬态特征与全局变化趋势的问题，提高了模型对微小泄漏信号的识别能力。
Smart Images
Figure CN122881979A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas pipeline leakage anomaly detection technology, and more specifically, to an intelligent diagnostic method and system for acoustic signals for digital twins of oil and gas pipeline networks. Background Technology
[0002] Oil and gas pipelines, as crucial infrastructure for transporting oil and gas resources, are characterized by long transmission distances, wide coverage, and complex operating environments. Their safe operation is of paramount importance for ensuring energy supply security and protecting the ecological environment. During pipeline operation, leaks not only cause energy losses and environmental pollution but can also trigger serious safety accidents such as fires and explosions. Therefore, conducting leak anomaly detection and intelligent diagnosis of oil and gas pipelines has significant engineering and application value for achieving safe pipeline operation and intelligent maintenance. With the development of digital twin technology, constructing a virtual-real mapping system for oil and gas pipelines to achieve real-time perception, dynamic analysis, and intelligent diagnosis of pipeline operating status has become an important development direction for intelligent pipeline monitoring.
[0003] Currently, leak detection methods based on acoustic signals are widely used in the field of oil and gas pipeline leak monitoring due to their advantages such as high sensitivity and fast response speed. With the rapid development of deep learning technology, researchers are gradually adopting one-dimensional convolutional neural networks for automatic feature extraction and anomaly identification of acoustic signals. Compared with traditional methods that rely on manual feature extraction, this can effectively improve the automation level and identification accuracy of leak detection.
[0004] However, existing deep learning-based acoustic signal detection methods still have certain shortcomings: on the one hand, traditional convolutional structures have limited receptive fields, making it difficult to simultaneously consider both local transient features and global trends in acoustic signals, resulting in insufficient multi-scale feature representation capabilities; on the other hand, as the number of network layers increases, shallow structural features are prone to information attenuation or loss during deep feature modeling, affecting the model's ability to identify minute leak features. Furthermore, existing methods have a low degree of integration with digital twin systems, making it difficult to achieve online intelligent diagnosis and leak alarm based on real-time data. Therefore, there is an urgent need for an intelligent acoustic signal diagnosis method that combines digital twin technology with improved deep learning models to improve the accuracy and real-time performance of oil and gas pipeline leak anomaly detection. Summary of the Invention
[0005] The technical problem to be solved by this invention is:
[0006] To address the shortcomings of existing acoustic leak detection methods for oil and gas pipelines in identifying minute leaks and the limited intelligent diagnostic capabilities of digital twin systems.
[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:
[0008] This invention provides an intelligent diagnostic method for acoustic signals in digital twins of oil and gas pipeline networks, comprising the following steps:
[0009] S100: Collect pipeline acoustic signals along the oil and gas pipeline during operation and upload them to the oil and gas pipeline digital twin system to construct a pipeline acoustic signal dataset.
[0010] S200: Perform data preprocessing on the pipeline acoustic signal dataset collected in step S100, and divide the pipeline acoustic signal into time series segments according to a preset time window to construct sample data for model training and testing.
[0011] S300. Construct a hierarchical multi-scale feature compensation network. Through hierarchical multi-scale feature modeling and cross-layer feature compensation strategies, perform feature modeling and leakage anomaly detection on acoustic signals. The hierarchical multi-scale feature compensation network includes an input module, a pyramid-shaped multi-scale feature extraction module, a selective dilation receptive field expansion module, a cross-layer channel alignment residual compensation module, and a classification output module.
[0012] The input module receives preprocessed pipe acoustic signal samples and inputs them into the network for feature extraction. The pyramid-shaped multi-scale feature extraction module constructs a hierarchical multi-scale feature representation using progressively increasing convolutional kernel sizes. The selective dilation receptive field expansion module further extracts features by introducing dilated convolutional structures. The cross-layer channel alignment residual compensation module performs channel alignment and residual fusion between shallow and deep features. The classification output module uses fully connected layers and a Softmax function to output the leakage anomaly category probability and leakage anomaly discrimination result.
[0013] S400. Divide the pipeline acoustic signal dataset into training samples and test samples. Use the training samples to train the hierarchical multi-scale feature compensation network constructed in step S300, and use the test samples to verify the model, thereby obtaining an intelligent diagnostic model for detecting leaks and anomalies in oil and gas pipelines.
[0014] S500: Embed the intelligent diagnostic model obtained in step S400 into the digital twin system, perform online analysis on the real-time collected pipeline acoustic signals, realize abnormal detection of oil and gas pipeline leaks, and output corresponding alarm information or leakage level when abnormal acoustic features are detected.
[0015] Further, in step S100, the pipe acoustic signal is a discrete-time sequence signal:
[0016] (1)
[0017] in: Let be the amplitude of the i-th group of sound wave signals at time t, and T be the number of sampling points of a group of sound wave signals;
[0018] The collected pipeline acoustic signals are uploaded to the oil and gas pipeline digital twin system, and the acoustic signal dataset D is composed of N sets of acoustic signals, represented as:
[0019] (2)
[0020] in, This represents the number of acoustic signal samples.
[0021] Further, in step S200, the data preprocessing includes denoising and normalization, wherein the normalization employs min-max normalization or Z-score normalization.
[0022] Mini-Max Normalization:
[0023] (3)
[0024] Z-score standardization:
[0025] (4)
[0026] in, The mean, Standard deviation Let be the value of the pipe acoustic signal at time t after minimum-maximum normalization or Z-score normalization, and min(x) and max(x) be the minimum and maximum values of the current pipe acoustic signal amplitude, respectively.
[0027] Further, in step S200, the acoustic signal is segmented into time series according to a preset time window, and samples are constructed using a sliding window with a window length of [missing information]. Step size is , No. Sample Represented as:
[0028] (5).
[0029] Further, in step S300,
[0030] The pyramid-shaped multi-scale feature extraction module uses one-dimensional convolution to model the features of the acoustic wave sequence. The convolution operation of a convolutional layer is represented as: :
[0031] (6)
[0032] in, For the first Layer convolution kernel length, For convolution kernel parameters, For bias terms, It is a non-linear activation function;
[0033] Convolution kernel size satisfies Furthermore, the receptive field is expanded through a hierarchical progression to effectively express the characteristics of acoustic signals at different time scales;
[0034] The progressively increasing kernel size is taken as: , and This corresponds to hierarchical modeling from local features to global features;
[0035] The selective dilation receptive field extension module introduces dilated convolutions in preset key convolutional layers. Shallow convolutional layers use standard convolutions to preserve local details, while key convolutional layers use dilation rates. Dilated convolution expands the receptive field; the operational form of dilated convolution is:
[0036] (7)
[0037] in, The expansion rate;
[0038] The selective expansion strategy satisfies:
[0039] (8)
[0040] in, Represented as the first The expansion rate of each convolutional layer A collection of local detail modeling layers. For the set of key receptive field extension layers;
[0041] The cross-layer channel alignment residual compensation module, through Convolution maps dimensions to shallow features:
[0042] (9)
[0043] in, This indicates the network layer number where the shallow features are located. These are shallow features. These are shallow features after channel alignment;
[0044] The shallow features after channel alignment are added and fused with the corresponding deep features element by element:
[0045] (10)
[0046] in, As a feature of fusion, This indicates the network layer number where the deep feature undergoing residual compensation resides, and , For deep features;
[0047] The classification output module uses a fully connected layer and a Softmax function to output the probability of leakage anomaly categories, and based on... Output leakage anomaly detection results ,in, This represents the predicted probability that the acoustic signal belongs to the c-th type of leakage anomaly.
[0048] Further, in step S400, pipeline acoustic signal samples under normal operating conditions and different leakage conditions are obtained from the cloud of the oil and gas pipeline digital twin system. A pipeline acoustic signal dataset is constructed from the collected pipeline acoustic signals, and the pipeline acoustic signal dataset is divided into training samples and test samples according to a preset ratio. The constructed training samples are input into the hierarchical multi-scale feature compensation network for training, and the model is verified using test samples to obtain an intelligent diagnostic model for detecting oil and gas pipeline leakage anomalies.
[0049] Furthermore, the oil and gas pipeline digital twin system includes an intelligent sensing layer, a data transmission layer, a digital twin modeling layer, and an intelligent analysis and application layer. The intelligent sensing layer includes an acoustic sensor unit, a data acquisition terminal, and a time synchronization module, used to acquire acoustic signals during the operation of the oil and gas pipeline in steps S100 and S200 and generate raw data with timestamps. The data transmission layer is used to upload the acoustic signals to the cloud or the pipeline control center. The digital twin modeling layer is used to establish the correlation mapping between pipeline topology, pipe segment attributes, operating parameters, and acoustic signals. The intelligent analysis and application layer is used to complete the pipeline acoustic signal feature modeling, leakage anomaly detection, and alarm in steps S300-S500.
[0050] An intelligent diagnostic system for acoustic signals of digital twins for oil and gas pipeline networks is provided. The system has program modules corresponding to the above steps and executes the steps in the above-described intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks during operation.
[0051] A computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement the steps of an intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks.
[0052] Compared with the prior art, the beneficial effects of the present invention are:
[0053] This invention combines deep learning technology with a digital twin system, introducing a hierarchical multi-scale feature compensation network to achieve intelligent analysis and leak anomaly detection of acoustic signals from oil and gas pipelines. Through the hierarchical multi-scale feature modeling structure, the network can extract feature information of acoustic signals at different time scales layer by layer, effectively solving the problems of limited receptive field and difficulty in simultaneously considering local transient features and global change trends in traditional CNNs, thus improving the model's ability to identify minute leak signals.
[0054] This invention expands the receptive field of the network without significantly increasing the number of model parameters by introducing a selective dilation structure into key convolutional layers, thereby enhancing the model's ability to model long-range correlated features and improving the accuracy of leak detection under complex operating conditions. Simultaneously, by constructing a cross-layer feature compensation mechanism, shallow structural features are fused with deep semantic features, effectively avoiding the loss of shallow information during deep network feature modeling and improving the model's feature representation ability and training stability. Finally, the trained intelligent diagnostic model is embedded into a digital twin system to achieve real-time online analysis and leak anomaly detection of acoustic signals from oil and gas pipelines. This enables the system to promptly output leak alarm information and leak type, improving the accuracy and real-time performance of pipeline leak detection, reducing economic losses, and enhancing the safety and intelligence level of oil and gas pipeline operation. Attached Figure Description
[0055] Figure 1 This is a flowchart of an intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks, as described in an embodiment of the present invention.
[0056] Figure 2 This is a structural diagram of the digital twin system for oil and gas pipelines in an embodiment of the present invention;
[0057] Figure 3 This is a diagram showing the hierarchical multi-scale feature compensation network structure and parameter settings in an embodiment of the present invention;
[0058] Figure 4 This is an iterative curve of the accuracy and loss value of the model on the training and validation sets in an embodiment of the present invention;
[0059] Figure 5 This is a diagram showing the confusion matrix of the model on the test set in an embodiment of the present invention. Detailed Implementation
[0060] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0061] Specific Implementation Plan 1: Combining Figures 1 to 3 As shown, this invention provides an intelligent diagnostic method for acoustic signals in digital twins of oil and gas pipeline networks, comprising the following steps:
[0062] S100. Acoustic wave signals during the operation of oil and gas pipelines are collected by acoustic wave sensors deployed along the pipeline and uploaded to the digital twin system of oil and gas pipelines to build a pipeline acoustic wave signal dataset.
[0063] The oil and gas pipeline digital twin system comprises an intelligent sensing layer, a data transmission layer, a digital twin modeling layer, and an intelligent analysis and application layer. The intelligent sensing layer includes an acoustic sensor unit, a data acquisition terminal, and a time synchronization module, used to collect acoustic signals during pipeline operation and generate timestamped raw data. The data transmission layer uploads the acoustic signals to the cloud or pipeline control center. The digital twin modeling layer establishes a mapping between pipeline topology, pipe segment attributes, operating parameters, and acoustic signals. The intelligent analysis and application layer performs acoustic signal feature modeling, leak anomaly detection, and alarm functions.
[0064] The acoustic signal from the pipeline is a discrete-time sequence signal, represented as follows:
[0065] (1)
[0066] in: Let be the amplitude of the i-th group of sound wave signals at time t, and T be the number of sampling points of a group of sound wave signals;
[0067] The collected pipeline acoustic signals are uploaded to the oil and gas pipeline digital twin system, and the acoustic signal dataset D is composed of N sets of acoustic signals, represented as:
[0068] (2)
[0069] in, The number of acoustic signal samples;
[0070] S200. Perform data preprocessing on the pipeline acoustic signal dataset collected in step S100, and divide the acoustic signal into time series segments according to a preset time window to construct sample data for model training and testing.
[0071] Specifically, including,
[0072] The data preprocessing includes noise reduction and normalization. Normalization employs either minimum-maximum normalization or Z-score normalization.
[0073] Mini-Max Normalization:
[0074] (3)
[0075] Z-score standardization:
[0076] (4)
[0077] in, The mean, Standard deviation Let be the value of the pipe acoustic signal at time t after minimum-maximum normalization or Z-score normalization, and min(x) and max(x) be the minimum and maximum values of the current pipe acoustic signal amplitude, respectively.
[0078] The acoustic signal is segmented into time series according to a preset time window, and samples are constructed using a sliding window with a window length of [value missing]. Step size is , No. Sample Represented as:
[0079] (5)
[0080] S300: Construct a hierarchical multi-scale feature compensation network, and perform feature modeling and leakage anomaly detection of acoustic signals through hierarchical multi-scale feature modeling and cross-layer feature compensation strategies;
[0081] Specifically, including,
[0082] The hierarchical multi-scale feature compensation network includes an input module, a pyramid-shaped multi-scale feature extraction module, a selective dilation receptive field expansion module, a cross-layer channel alignment residual compensation module, and a classification output module. The hierarchical multi-scale feature compensation network is used to perform feature modeling on pipeline acoustic signals and to achieve leakage anomaly detection.
[0083] The pyramid-shaped multi-scale feature extraction module uses one-dimensional convolution to model the features of the acoustic wave sequence. The convolution operation of a convolutional layer is represented as: :
[0084] (6)
[0085] in, For the first Layer convolution kernel length, For convolution kernel parameters, For bias terms, It is a non-linear activation function;
[0086] The pyramid-shaped multi-scale feature extraction module constructs hierarchical multi-scale feature representations using progressively increasing convolutional kernel sizes, with the kernel sizes satisfying... Furthermore, the receptive field is expanded through a hierarchical progression to effectively express the characteristics of acoustic signals at different time scales;
[0087] The progressively increasing kernel size is taken as: , and This corresponds to hierarchical modeling from local features to global features;
[0088] The selective dilation receptive field extension module introduces dilated convolutions only in preset key convolutional layers. Shallow convolutional layers use standard convolutions to preserve local details, while key convolutional layers use dilation rates. Dilated convolution expands the receptive field; the operational form of dilated convolution is:
[0089] (7)
[0090] in, The dilation rate is used to expand the receptive field without increasing the number of convolution kernel parameters, thereby enhancing the ability to model long-range correlated features.
[0091] The selective expansion strategy satisfies:
[0092] (8)
[0093] in, Represented as the first The expansion rate of each convolutional layer A collection of local detail modeling layers. This is a set of key receptive field extension layers to avoid the grid effect caused by global expansion stacking and reduce redundant computation;
[0094] The cross-layer channel alignment residual compensation module performs channel alignment and residual fusion on shallow and deep features to compensate for and enhance the shallow structural information on the deep semantic features, thereby improving the stability and discrimination capability of acoustic signal feature modeling; including,
[0095] pass Convolution maps the dimensions of shallow features to match the dimensions of their corresponding deep features. This mapping relationship is expressed as follows:
[0096] (9)
[0097] in, This indicates the network layer number where the shallow features are located. These are shallow features. These are shallow features after channel alignment;
[0098] The shallow features after channel alignment are fused with the corresponding deep features element-wise by addition. The fusion relationship is expressed as follows:
[0099] (10)
[0100] in, As a feature of fusion, This indicates the network layer number where the deep feature undergoing residual compensation resides, and , For deep features;
[0101] The classification output module uses a fully connected layer and a Softmax function to output the probability of leakage anomaly categories, and based on... Output leakage anomaly detection results ,in, This represents the predicted probability that the acoustic signal belongs to the c-th type of leakage anomaly.
[0102] S400. Divide the pipeline acoustic signal dataset into training samples and test samples. Use the training samples to train the hierarchical multi-scale feature compensation network constructed in step S300, and use the test samples to verify the model, thereby obtaining an intelligent diagnostic model for detecting leaks and anomalies in oil and gas pipelines.
[0103] The system acquires pipeline acoustic signal samples under normal operating conditions and different leakage conditions from the cloud of the oil and gas pipeline digital twin system. It constructs a pipeline acoustic signal dataset from the collected pipeline acoustic signals and divides the dataset into training samples and test samples according to a preset ratio. The constructed training samples are input into the hierarchical multi-scale feature compensation network for training, and the model is validated using test samples to obtain an intelligent diagnostic model for detecting oil and gas pipeline leakage anomalies.
[0104] S500: Embed the intelligent diagnostic model obtained in step S400 into the digital twin system, perform online analysis on the real-time collected pipeline acoustic signals, realize abnormal detection of oil and gas pipeline leaks, and output corresponding alarm information or leakage level when abnormal acoustic features are detected.
[0105] Specific Implementation Scheme 2: The present invention provides an intelligent diagnostic system for acoustic signals of digital twins of oil and gas pipeline networks. This system has program modules corresponding to the above steps, and executes the steps in the above-mentioned intelligent diagnostic method for acoustic signals of digital twins of oil and gas pipeline networks when running.
[0106] The other combinations and connections in this implementation scheme are the same as in Specific Implementation Scheme 1.
[0107] Specific Implementation Scheme 3: The present invention provides a computer-readable storage medium storing a computer program configured to implement, when called by a processor, the steps of an intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks.
[0108] The other combinations and connections in this implementation scheme are the same as in Specific Implementation Scheme 1.
[0109] Example
[0110] In step S100, acoustic sensor units deployed at key nodes along the oil and gas pipeline collect acoustic signal parameters during pipeline operation in real time. These acoustic sensor units, through a data acquisition terminal and signal transmission module, upload the collected raw acoustic signals to the cloud platform of the oil and gas pipeline digital twin system for centralized storage and management. The digital twin system establishes a mapping relationship between the physical entity of the pipeline and its virtual model, enabling real-time synchronous monitoring and data mapping of the pipeline's operating status. This constructs a pipeline acoustic signal dataset, providing data support for subsequent leak anomaly detection model training. In this embodiment, pipeline acoustic signal samples under different operating conditions (normal, small leak, large leak) are extracted from the digital twin system. 2000 samples are collected for each condition, with 784 sampling points per sample, forming a multi-condition acoustic signal dataset for subsequent model training and verification.
[0111] In step S200, the pipe acoustic signal data collected in step S100 is preprocessed to eliminate the influence of different dimensions and amplitude ranges on model training. Specifically, this includes: First, normalizing the original acoustic signal to map the signal amplitude to a preset interval to improve the stability and convergence speed of model training; Second, based on the one-dimensional time series characteristics of the acoustic signal, segmenting the acoustic signal according to a preset time window length to construct a data sample of fixed length 784, labeled according to category labels, where label 0 represents a large leakage state, label 1 represents a small leakage state, and label 2 represents a normal operating state; Finally, the constructed dataset is divided into a training set, a validation set, and a test set according to a preset ratio. Of the 2000 samples in each operating condition, the first 1800 are used as training candidate samples, the last 200 are used as test samples, and 10% of the training candidate samples are randomly allocated as validation samples to ensure a balanced distribution of categories between the training set and the validation set.
[0112] In step S300, a hierarchical multi-scale feature compensation network is constructed to perform feature modeling and leakage anomaly detection on the pipeline acoustic signal. The hierarchical multi-scale feature compensation network includes an input module, a pyramid-shaped multi-scale feature extraction module, a selective dilation receptive field expansion module, a cross-layer channel alignment residual compensation module, and a classification output module. The specific network structure model and parameter settings are as follows: Figure 3As shown in the diagram, the input module receives preprocessed pipeline acoustic signal samples and inputs them into the network for feature extraction. The pyramid-shaped multi-scale feature extraction module consists of two one-dimensional convolutional structures with kernel sizes of 8 and 16, and output channels of 8 and 16, respectively. The selective dilation receptive field expansion module further extracts features by introducing a dilated convolutional structure with a kernel size of 32 and a dilation coefficient of 2. The cross-layer channel alignment residual compensation module maps the number of channels of shallow features to match the number of channels of deep features using a 1×1 convolutional structure, and fuses the shallow and deep features element-wise to achieve cross-layer feature compensation, thereby enhancing the network's ability to express features at different scales and improving model stability. The classification output module consists of a flat layer and a fully connected layer, which maps the fused features to the classification results of various operating conditions, outputs the model's predicted probability for different pipeline operating states, and realizes pipeline leakage anomaly detection.
[0113] In step S400, the training set is input into a hierarchical multi-scale feature compensation network for training. The cross-entropy loss function is used as the objective function for model optimization, and the Adam optimization algorithm is used to iteratively update the network parameters to minimize the model's prediction error. During model training, the validation set is used to evaluate the model's performance in real time. By monitoring the trends of accuracy and loss values on the training and validation sets, the model's training status is dynamically monitored to prevent overfitting and improve its generalization ability. The iterative curves of the model's accuracy and loss values on the training and validation sets are shown below. Figure 4 As shown in the figure, the model loss value gradually decreases and the accuracy gradually increases and tends to stabilize as the number of training iterations increases, indicating that the constructed model has good convergence performance and stability.
[0114] After model training is complete, the test set is input into the trained model for performance validation. The model outputs classification results under different operating conditions, and the confusion matrix is used to evaluate the model's classification performance. The confusion matrix results on the test set are as follows: Figure 5 As shown, the diagonal elements of the confusion matrix represent the number of correctly classified samples. The confusion matrix shows that the model has high classification accuracy and good leakage anomaly identification ability.
[0115] Depend on Figure 5As shown in the confusion matrix and Table 1 below, the proposed hierarchical multi-scale feature compensation network correctly identified 597 samples on the test set, achieving an overall classification accuracy of 99.50% and an error rate of only 0.50%. Meanwhile, the model's average precision, average recall, and average F1 score reached 99.50%, 99.50%, and 99.50%, respectively, as shown in Table 1. This indicates that the proposed method can effectively identify the three operating states of normal, small leakage, and large leakage, demonstrating high classification accuracy and good generalization ability.
[0116] Table 1 Results of each evaluation indicator
[0117]
[0118] In step S500, the trained leak anomaly diagnosis model is embedded into the digital twin system of the oil and gas pipeline to achieve online intelligent analysis of the real-time acquired acoustic signals. When the digital twin system receives the real-time acoustic signal collected by the acoustic sensor, it first preprocesses the signal and constructs model input data. Then, the processed acoustic signal is input into the leak anomaly diagnosis model for feature extraction and state discrimination. When the model detects abnormal acoustic features, the digital twin system automatically outputs the corresponding abnormal alarm information and determines the leakage level based on the intensity of the abnormal features and the classification results, thereby achieving real-time monitoring and early warning of oil and gas pipeline leak anomalies. Through the above method, real-time intelligent analysis of acoustic signals and leak anomaly detection of oil and gas pipelines are realized, improving the accuracy, real-time performance, and intelligence level of oil and gas pipeline leak detection.
[0119] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.
Claims
1. A method for intelligent diagnosis of acoustic signals for digital twins of oil and gas pipeline networks, characterized in that, Includes the following steps: S100: Collect pipeline acoustic signals along the oil and gas pipeline during operation and upload them to the oil and gas pipeline digital twin system to build a pipeline acoustic signal dataset. S200: Perform data preprocessing on the pipeline acoustic signal dataset collected in step S100, and divide the pipeline acoustic signal into time series segments according to a preset time window to construct sample data for model training and testing. S300. Construct a hierarchical multi-scale feature compensation network. Through hierarchical multi-scale feature modeling and cross-layer feature compensation strategies, perform feature modeling and leakage anomaly detection on acoustic signals. The hierarchical multi-scale feature compensation network includes an input module, a pyramid-shaped multi-scale feature extraction module, a selective dilation receptive field expansion module, a cross-layer channel alignment residual compensation module, and a classification output module. The input module receives preprocessed pipe acoustic signal samples and inputs them into the network for feature extraction. The pyramid-shaped multi-scale feature extraction module constructs a hierarchical multi-scale feature representation using progressively increasing convolutional kernel sizes. The selective dilation receptive field expansion module further extracts features by introducing dilated convolutional structures. The cross-layer channel alignment residual compensation module performs channel alignment and residual fusion between shallow and deep features. The classification output module uses fully connected layers and a Softmax function to output the leakage anomaly category probability and leakage anomaly discrimination result. S400. Divide the pipeline acoustic signal dataset into training samples and test samples. Use the training samples to train the hierarchical multi-scale feature compensation network constructed in step S300, and use the test samples to verify the model, thereby obtaining an intelligent diagnostic model for detecting leaks and anomalies in oil and gas pipelines. S500: Embed the intelligent diagnostic model obtained in step S400 into the digital twin system, perform online analysis on the real-time collected pipeline acoustic signals, realize abnormal detection of oil and gas pipeline leaks, and output corresponding alarm information or leakage level when abnormal acoustic features are detected.
2. The intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks according to claim 1, characterized in that: In step S100, the pipe acoustic signal is a discrete-time sequence signal: （1） in: Let be the amplitude of the i-th group of sound wave signals at time t, and T be the number of sampling points of a group of sound wave signals; The collected pipeline acoustic signals are uploaded to the oil and gas pipeline digital twin system, and the N sets of acoustic signals constitute an acoustic signal dataset D, represented as: （2） in, This represents the number of acoustic signal samples.
3. The method for intelligent diagnosis of acoustic signals for digital twins of oil and gas pipeline networks according to claim 2, characterized in that: In step S200, the data preprocessing includes denoising and normalization. The normalization process employs min-max normalization or Z-score normalization. Mini-Max Normalization: （3） Z-score standardization: （4） in, The mean, Standard deviation Let be the value of the pipe acoustic signal at time t after minimum-maximum normalization or Z-score normalization, and min(x) and max(x) be the minimum and maximum values of the current pipe acoustic signal amplitude, respectively.
4. The intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks according to claim 3, characterized in that: In step S200, the acoustic signal is segmented into time series according to a preset time window, and samples are constructed using a sliding window with a window length of [missing information]. Step size is , No. Sample Represented as: （5）。 5. The method for intelligent diagnosis of acoustic signals for digital twins of oil and gas pipeline networks according to claim 4, characterized in that: In step S300, The pyramid-shaped multi-scale feature extraction module uses one-dimensional convolution to model the features of the acoustic wave sequence. The convolution operation of a convolutional layer is represented as: : （6） in, For the first Layer convolution kernel length, For convolution kernel parameters, For bias terms, It is a non-linear activation function; Convolution kernel size satisfies Furthermore, the receptive field is expanded through a hierarchical progression to effectively express the characteristics of acoustic signals at different time scales; The progressively increasing kernel size is taken as: , and This corresponds to hierarchical modeling from local features to global features; The selective dilation receptive field extension module introduces dilated convolutions in preset key convolutional layers. Shallow convolutional layers use standard convolutions to preserve local details, while key convolutional layers use dilation rates. Dilated convolution expands the receptive field; the operational form of dilated convolution is: （7） in, The expansion rate; The selective expansion strategy satisfies: （8） in, Represented as the first The dilation rate of each convolutional layer A collection of local detail modeling layers. For key receptive field extension layers; The cross-layer channel alignment residual compensation module, through Convolution maps dimensions to shallow features: （9） in, This indicates the network layer number where the shallow features are located. These are shallow features. These are shallow features after channel alignment; The shallow features after channel alignment are added and fused with the corresponding deep features element by element: （10） in, As a feature of fusion, This indicates the network layer number where the deep feature undergoing residual compensation resides, and , For deep features; The classification output module uses a fully connected layer and a Softmax function to output the probability of leakage anomaly categories, and based on... Output leakage anomaly detection results ,in, This represents the predicted probability that the acoustic signal belongs to the c-th type of leakage anomaly.
6. The method for intelligent diagnosis of acoustic signals for digital twins of oil and gas pipeline networks according to claim 5, characterized in that: In step S400, pipeline acoustic signal samples under normal operating conditions and different leakage conditions are obtained from the cloud of the oil and gas pipeline digital twin system. A pipeline acoustic signal dataset is constructed from the collected pipeline acoustic signals, and the pipeline acoustic signal dataset is divided into training samples and test samples according to a preset ratio. The constructed training samples are input into the hierarchical multi-scale feature compensation network for training, and the model is verified using test samples to obtain an intelligent diagnostic model for detecting oil and gas pipeline leakage anomalies.
7. The intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks according to claim 6, characterized in that: The oil and gas pipeline digital twin system includes an intelligent sensing layer, a data transmission layer, a digital twin modeling layer, and an intelligent analysis and application layer. The intelligent sensing layer includes an acoustic sensor unit, a data acquisition terminal, and a time synchronization module, used to acquire acoustic signals during the operation of the oil and gas pipeline in steps S100 and S200 and generate raw data with timestamps. The data transmission layer is used to upload the acoustic signals to the cloud or the pipeline control center. The digital twin modeling layer is used to establish the correlation mapping between pipeline topology, pipe segment attributes, operating parameters, and acoustic signals. The intelligent analysis and application layer is used to complete the pipeline acoustic signal feature modeling, leakage anomaly detection, and alarm in steps S300-S500.
8. An intelligent diagnostic system for acoustic signals in digital twins of oil and gas pipeline networks, characterized in that: The system has a program module corresponding to the steps of any one of the claims 1-7 above, and executes the steps in the above-described intelligent diagnostic method for acoustic signals of digital twins for oil and gas pipeline networks when it is run.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the intelligent acoustic signal diagnostic method for digital twins of oil and gas pipeline networks as described in any one of claims 1-7.