An intelligent identification method and system for oil and gas pipeline threat events
By combining distributed fiber acoustic sensing technology and the Transformer-TCN-SPP model, the problems of low identification accuracy and complex data processing in oil and gas pipeline monitoring are solved, and efficient and accurate threat event identification is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-28
- Publication Date
- 2026-04-10
AI Technical Summary
Existing oil and gas pipeline monitoring methods suffer from low accuracy, complex operation and maintenance, and high costs. Furthermore, fiber optic distributed acoustic sensing technology suffers from severe data redundancy in data processing, resulting in low accuracy in threat event identification and frequent false alarms and missed alarms.
Distributed fiber optic acoustic sensing technology is used to collect vibration signals from oil and gas pipelines. After denoising and standardization preprocessing, global features are extracted using a Transformer model, and multi-scale local features are extracted using a TCN-SPP model. Finally, threat events are classified using a multilayer perceptron.
It significantly improves the accuracy and response speed of threat event detection, enhances the accuracy, robustness and generalization ability of the model, and meets the needs of real-time threat detection.
Smart Images
Figure CN120632560B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of oil pipeline safety monitoring, and in particular to an oil and gas pipeline threat event intelligent identification method and system. BACKGROUND
[0002] Traditional pipeline monitoring methods such as negative pressure wave method, stress wave method and mass balance method have the problems of low identification accuracy, complex operation and maintenance, high cost, etc. These methods usually rely on relatively simple physical principles to monitor the abnormal conditions inside the pipeline, but due to environmental factors, these methods have limitations in precision and sensitivity. For example, the negative pressure wave method may not react in time when there is a small leak in the pipeline, and the stress wave method is easily disturbed by external interference, leading to misjudgment.
[0003] Although the application of optical fiber distributed acoustic sensing (DAS) technology greatly improves the sensitivity of pipeline threat detection, it can realize real-time monitoring along the pipeline, but it still faces many challenges in practical application. First, the amount of data obtained by DAS technology is huge, and processing these data requires strong computing power and efficient algorithm support, however, existing data processing technologies often cannot fully extract effective information, resulting in serious redundancy and increasing the complexity of data analysis. Second, the recognition accuracy of threat events is still not high, and frequent false positives and false negatives not only affect the reliability of monitoring, but also cause trouble for subsequent operation and decision-making. SUMMARY
[0004] To solve the above problems in the prior art, the present application provides an oil and gas pipeline threat event intelligent identification method and system. The application first uses distributed optical fiber acoustic sensing technology to collect oil and gas pipeline vibration signals; then, the vibration signals are denoised and standardized for preprocessing; secondly, the preprocessed signals are input into the Transformer model to extract global features of the signals; at the same time, the preprocessed signals are decomposed into short-term signals to obtain multiple effective short-term high-energy period signals, which are sent to the TCN-SPP model for multi-scale local feature extraction; thirdly, the global features and short-term signal local features are sent to the splicing layer for feature fusion to obtain a comprehensive feature vector; finally, the fused feature vector is sent to the multilayer perceptron for pipeline event classification to output the event type prediction result. The application effectively improves the detection accuracy and response speed of threat events, and enhances the accuracy, robustness and generalization ability of the model. To achieve the above purpose, the technical solution is as follows:
[0005] On the one hand, the present application provides an oil and gas pipeline threat event intelligent identification method, which comprises:
[0006] S1, distribute optical fiber sensors on the oil and gas pipeline, use optical fiber acoustic sensing technology to obtain oil and gas pipeline vibration signals;
[0007] S2, denoising and standardizing the oil and gas pipeline vibration signals to obtain preprocessed signals;
[0008] S3, according to the preprocessed signals, extracting global features through a Transformer model to obtain a global feature dataset;
[0009] S4, decomposing the preprocessed signals into short-term high-energy periodic signals;
[0010] S5, according to the short-term high-energy periodic signals, extracting multi-scale local features through a TCN-SPP model to obtain a multi-scale local feature dataset;
[0011] S6, fusing the global feature dataset and the multi-scale local feature dataset to obtain a comprehensive feature vector;
[0012] S7, according to the comprehensive feature vector, classifying threat events through a multilayer perceptron (MLP) to obtain a pipeline threat event type prediction result.
[0013] Optionally, in S2, the oil and gas pipeline vibration signals are denoised and standardized to obtain preprocessed signals, including:
[0014] S21, removing noise from the oil and gas pipeline vibration signals to obtain denoised oil and gas pipeline vibration signals;
[0015] S22, according to the denoised oil and gas pipeline vibration signals, performing standardization processing to unify the scale to obtain preprocessed signals.
[0016] Optionally, in S4, the preprocessed signals are decomposed into short-term high-energy periodic signals, including:
[0017] S41, according to the preprocessed signals, dividing them according to a predetermined sliding window size to obtain local short-term signal segments;
[0018] S42, according to the local short-term signal segments, obtaining the energy intensity of the local short-term signal segments through formula (1),
[0019]
[0020] wherein E represents energy intensity, N represents the total number of signal sample points, x n represents the signal amplitude at the nth time step;
[0021] S43, according to the local short-term signal segment, through frequency domain analysis recognition, get the periodicity characteristics of the local short-term signal segment;
[0022] S44, according to the energy intensity of the local short-term signal segment, through formula (2), get the signal-to-noise ratio of the local short-term signal segment,
[0023]
[0024] In the formula, SNR represents the signal-to-noise ratio, E signal represents the energy intensity of the local short-term signal segment, E noise represents the noise intensity of the local short-term signal segment;
[0025] S45, according to the signal-to-noise ratio of the local short-term signal segment, set the signal-to-noise ratio threshold, get the effective local short-term signal segment;
[0026] S46, according to the effective local short-term signal segment and the periodicity characteristics of the local short-term signal segment, through analysis screening, get the short-term high-energy periodic signal.
[0027] Optionally, S43 according to the local short-term signal segment, through frequency domain analysis recognition, get the periodicity characteristics of the local short-term signal segment, including:
[0028] S431, according to the local short-term signal segment, through formula (3), Fourier transform, get the local short-term signal in frequency domain,
[0029]
[0030] Where, F(ω) represents the local short-term signal in frequency domain, f(t) represents the local short-term signal segment, ω represents the angular frequency, e -iwt represents the complex exponential function;
[0031] S432, according to the local short-term signal in frequency domain, through detection of significant peak value to identify periodic behavior, output frequency domain peak value, get the periodicity characteristics of the local short-term signal segment.
[0032] Optionally, S5 according to the short-term high-energy periodic signal, through TCN-SPP model extraction multi-scale local feature, get multi-scale local feature dataset, including:
[0033] S51, according to the short-term high-energy periodic signal, adopt multi-layer time series convolution network TCN model, through multi-layer time series convolution network TCN processing, get multi-scale local signal feature map;
[0034] S52, according to the multi-scale local signal feature map, through SPP model processing, get multi-scale local feature dataset.
[0035] Optionally, the multi-layer time sequence convolution network TCN model adopts a specific dilated rate configuration.
[0036] Optionally, in S51, according to the short-term high-energy periodic signal, a multi-layer time sequence convolution network TCN model is adopted to obtain a multi-scale local signal feature map through multi-layer time sequence convolution network TCN processing, including:
[0037] S511, according to the short-term high-energy periodic signal, the time dependence of the short-term high-energy periodic signal is captured through causal convolution and dilated convolution, and the features of the short-term high-energy periodic signal at different time scales are obtained.
[0038] S512, according to the features of the short-term high-energy periodic signal at different time scales, a multi-scale local signal feature map is obtained by adjusting the dilated factor.
[0039] Optionally, in S52, according to the multi-scale local signal feature map, a multi-scale local feature dataset is obtained through SPP model processing, including:
[0040] S521, according to the multi-scale local signal feature map, a multi-scale pooling region is obtained through spatial pyramid pooling processing.
[0041] S522, according to the multi-scale pooling region, a multi-scale local feature dataset is obtained through maximum pooling or average pooling.
[0042] On the other hand, the present application provides an oil and gas pipeline threat event intelligent identification system, which is applied to an oil and gas pipeline threat event intelligent identification method, and the system includes:
[0043] A vibration signal acquisition module is used to distribute optical fiber sensors on an oil and gas pipeline, and an optical fiber acoustic sensing technology is used to obtain an oil and gas pipeline vibration signal.
[0044] A preprocessing module is used to denoise and standardize the preprocessing of the oil and gas pipeline vibration signal to obtain a preprocessing signal.
[0045] A global feature extraction module is used to extract global features through a Transformer model according to the preprocessing signal to obtain a global feature dataset.
[0046] A short-term signal decomposition module is used to perform short-term signal decomposition on the preprocessing signal to obtain a short-term high-energy periodic signal.
[0047] A multi-scale local feature extraction module is used to extract multi-scale local features through a TCN-SPP model according to the short-term high-energy periodic signal to obtain a multi-scale local feature dataset.
[0048] a feature fusion module configured to fuse the global feature dataset and the multi-scale local feature dataset to obtain a comprehensive feature vector;
[0049] a threat event classification module configured to classify the threat event according to the comprehensive feature vector through a multi-layer perceptron (MLP) to obtain a pipeline threat event type prediction result.
[0050] Compared with the prior art, the technical solution of the present application has at least the following beneficial effects:
[0051] On the one hand, the Transformer-TCN-SPP model for joint extraction of global and local features is adopted, the global time trend feature is extracted through the Transformer module, and the local significant amplitude feature is extracted through the TCN-SPP module, so that the model can perceive the overall trend and local mutation of the signal at the same time, effectively making up for the limitations of single feature extraction; on the second aspect, the multi-scale pyramid pooling (SPP) is introduced, which further enhances the perception ability of the model to different scale features, so that it can capture long-term and short-term key information at the same time, effectively improving the discrimination ability of complex events; on the third aspect, the TCN structure with a specific dilation rate is adopted, which takes into account long-period and short-period feature extraction, optimizes the feature receptive field distribution, and significantly improves the threat event recognition accuracy; on the fourth aspect, through reasonable module combination and parameter optimization, the model realizes the dual optimization of inference speed and model size while ensuring high accuracy, meeting the real-time threat detection requirements. BRIEF DESCRIPTION OF DRAWINGS
[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.
[0053] Figure 1 is a flowchart of an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0054] Figure 2 is four signal modes of the oil and gas pipeline vibration signal in the oil and gas pipeline threat event intelligent identification method embodiment of the present application, wherein figure (a) is a background noise signal mode, figure (b) is a manual operation signal mode, figure (c) is a vehicle passing signal mode, and figure (d) is a mechanical excavation signal mode;
[0055] Figure 3 is a flowchart for obtaining a preprocessed signal in the oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0056] Figure 4 is a preprocessed signal effect diagram in an oil and gas pipeline threat event intelligent identification method embodiment of the present application, wherein figure (a) is a preprocessed vehicle passing signal mode, and figure (b) is a preprocessed mechanical excavation signal mode;
[0057] Figure 5 is a structure schematic diagram of extracting global features in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0058] Figure 6 is a flowchart of obtaining a short-term high-energy periodic signal in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0059] Figure 7 is a flowchart of obtaining periodic characteristics of a local short-term signal segment in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0060] Figure 8 is a flowchart of obtaining a multi-scale local feature dataset in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0061] Figure 9 is a flow of obtaining a multi-scale local signal feature map in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0062] Figure 10 is a flowchart of obtaining a multi-scale local feature dataset according to the multi-scale local signal feature map in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0063] Figure 11 is a curve diagram of the average accuracy and loss of a Transformer-TCN-SPP model in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0064] Figure 12 is a comparison diagram of a confusion matrix in pipeline event classification in an oil and gas pipeline threat event intelligent identification method embodiment of the present application;
[0065] Figure 13 is a system block diagram of an oil and gas pipeline threat event intelligent identification system embodiment of the present application. DETAILED DESCRIPTION
[0066] The technical solutions in the present application will be described below with reference to the drawings.
[0067] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.
[0068] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.
[0069] like Figure 1 The flowchart shown is an embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention. The present invention provides an intelligent identification method for oil and gas pipeline threat events, which is implemented by an intelligent identification system for oil and gas pipeline threat events. The method includes:
[0070] S1. Distribute fiber optic sensors on oil and gas pipelines and use fiber optic acoustic sensing technology to obtain vibration signals of oil and gas pipelines.
[0071] Specifically, Figure 2 The embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention shows four signal modes of oil and gas pipeline vibration signals. Figure (a) shows the background noise signal mode, Figure (b) shows the manual operation signal mode, Figure (c) shows the vehicle passage signal mode, and Figure (d) shows the mechanical excavation signal mode. This embodiment collects four types of oil and gas pipeline vibration signals from the oil and gas pipeline system based on DAS technology: background noise, vehicle passage, vibration caused by manual knocking or excavation, and vibration caused by mechanical equipment (such as tillers or excavators). Figure 2 The signal patterns of four pipeline events were demonstrated: background noise manifested as low amplitude, high frequency, and stable vibration; the signal from manual knocking or digging had asymmetrical high amplitude spikes and decayed rapidly; although the signal amplitude of vehicle passage was greater than that of mechanical digging, the two signals were very similar and difficult to distinguish. Compared with background noise and vehicle passage, manual knocking or digging and mechanical digging posed a potential threat to pipeline integrity and safety, and therefore these two events required special attention.
[0072] S2. Denoise and standardize the vibration signal of the oil and gas pipeline to obtain the preprocessed signal;
[0073] Specifically, such as Figure 3 The flowchart shown in the embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention illustrates the process of obtaining a preprocessed signal. In step S2, the oil and gas pipeline vibration signal undergoes denoising and standardization preprocessing to obtain the preprocessed signal, including:
[0074] S21. The vibration signal of the oil and gas pipeline is subjected to noise removal processing to obtain the noise-removed vibration signal of the oil and gas pipeline.
[0075] Furthermore, such as Figure 4 The diagram shows the effect of preprocessed signals in an embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention. Figure (a) shows the preprocessed vehicle passage signal pattern, and Figure (b) shows the preprocessed mechanical excavation signal pattern. The vibration signal of the oil and gas pipeline is subjected to noise removal processing. In this embodiment, a median filtering-based denoising process is designed to address the high noise characteristics of the original DAS signal. This method can remove the fault noise interference generated by the engine during vehicle and mechanical operation in this embodiment, while not significantly changing the overall trend of the signal, ensuring the optimization of the denoised signal quality, so that the denoised signal can be used for subsequent processing and analysis.
[0076] S22. Based on the noise-reduced oil and gas pipeline vibration signal, standardize the signal to unify the scale and obtain the preprocessed signal.
[0077] Furthermore, the preprocessed signals are collectively referred to as the preprocessed signal dataset, which is divided into two parts: a training set and a test set, with a ratio of 7:3.
[0078] S3. Based on the preprocessed signal, extract global features using the Transformer model to obtain the global feature dataset;
[0079] Specifically, such as Figure 5 The schematic diagram shown in the embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention is a structural diagram for extracting global features. The structure for extracting global features is composed of a Transformer model, which mainly includes three core components: position encoding, multi-head self-attention mechanism, and feedforward network. In the entire Transformer model, the self-attention mechanism enables each position to pay attention to the information of all other positions without considering the length limit of the input sequence. This allows the Transformer model to efficiently capture long-term dependencies, making it particularly suitable for processing long-term sequence data. This enables the extraction of global features from DAS signals, thereby improving the accuracy of threat event identification and accelerating the training process, breaking through the efficiency bottleneck faced by traditional methods when processing long-sequence data.
[0080] S4. Perform short-term signal decomposition on the preprocessed signal to obtain a short-term high-energy periodic signal;
[0081] Specifically, such as Figure 6The flowchart for obtaining the short-term high-energy periodic signal in the oil and gas pipeline threat event intelligent identification method embodiment of the application shown in the figure, the preprocessed signal is decomposed into a short-term signal in S4, and a short-term high-energy periodic signal is obtained, including:
[0082] S41, according to the preprocessed signal, the local short-term signal segment is obtained by dividing according to a predetermined sliding window size;
[0083] S42, according to the local short-term signal segment, the energy intensity of the local short-term signal segment is obtained through formula (1),
[0084]
[0085] In the formula, E represents the energy intensity, N represents the total number of signal sample points, x n represents the signal amplitude of the nth time step;
[0086] S43, according to the local short-term signal segment, the periodicity characteristics of the local short-term signal segment are obtained through frequency domain analysis identification;
[0087] Further, as shown in the flowchart for obtaining the periodicity characteristics of the local short-term signal segment in the oil and gas pipeline threat event intelligent identification method embodiment of the application, Figure 7 S43, according to the local short-term signal segment, the periodicity characteristics of the local short-term signal segment are obtained through frequency domain analysis identification, including:
[0088] S431, according to the local short-term signal segment, the local short-term signal in the frequency domain is obtained through Fourier transform through formula (3),
[0089]
[0090] In the formula, F(ω) represents the local short-term signal in the frequency domain, f(t) represents the local short-term signal segment, ω represents the angular frequency, e -iwt represents the complex exponential function;
[0091] S432, according to the local short-term signal in the frequency domain, the periodic behavior is identified by detecting the significant peak value, the frequency domain peak value is output, and the periodicity characteristics of the local short-term signal segment are obtained.
[0092] S44, according to the energy intensity of the local short-term signal segment, the signal-to-noise ratio of the local short-term signal segment is obtained through formula (2),
[0093]
[0094] In the formula, SNR represents the signal-to-noise ratio, E signal represents the energy intensity of the local short-term signal segment, Enoise Indicates the noise intensity of a local short-term signal segment;
[0095] S45. Based on the signal-to-noise ratio of the local short-term signal segment, set a signal-to-noise ratio threshold to obtain an effective local short-term signal segment;
[0096] S46. Based on the effective local short-term signal segment and the periodic characteristics of the local short-term signal segment, a short-term high-energy periodic signal is obtained through analysis and screening.
[0097] S5. Based on the short-term high-energy periodic signal, extract multi-scale local features using the TCN-SPP model to obtain a multi-scale local feature dataset;
[0098] Specifically, such as Figure 8 The flowchart shown in the embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention illustrates the process of obtaining a multi-scale local feature dataset. In step S5, based on the short-term high-energy periodic signal, multi-scale local features are extracted using the TCN-SPP model to obtain the multi-scale local feature dataset, which includes:
[0099] S51. Based on the short-term high-energy periodic signal, a multi-layer temporal convolutional network (TCN) model is adopted. Through the processing of the multi-layer temporal convolutional network (TCN), a multi-scale local signal feature map is obtained.
[0100] Furthermore, such as Figure 9 The flowchart of the intelligent identification method for oil and gas pipeline threat events of the present invention, shown in embodiment S51, describes the process of obtaining multi-scale local signal feature maps. Based on the short-term high-energy periodic signal, a multi-layer temporal convolutional network (TCN) model is used to process the multi-scale local signal feature map, including:
[0101] S511. Based on the short-term high-energy periodic signal, capture the time dependence of the short-term high-energy periodic signal through causal convolution and dilated convolution to obtain the characteristics of the short-term high-energy periodic signal at different time scales.
[0102] S512. Based on the characteristics of the short-term high-energy periodic signal at different time scales, multi-scale local signal feature maps are obtained by adjusting the expansion factor.
[0103] Furthermore, this multi-layer temporal convolutional network (TCN) model employs a specific dilation rate configuration;
[0104] Furthermore, the multi-layer temporal convolutional network (TCN) model was obtained through adjustment experiments. The experimental results of the dilation rate adjustment experiment are shown in Table 1 below. The experimental results show that the model has the highest recognition speed and recognition accuracy when the specific dilation rate combination configuration of 6, 4, 2, 2 is used.
[0105] Table 1. Results of the expansion rate adjustment experiment.
[0106] TCN1 TCN2 TCN3 TCN4 Accuracy 6 4 2 2 0.992 2 2 4 6 0.972 8 6 4 2 0.976 2 2 2 2 0.944 6 4 4 2 0.959 6 4 2 4 0.988
[0107] S52. Based on the multi-scale local signal feature map, the multi-scale local feature dataset is obtained by processing it through the SPP model.
[0108] Furthermore, such as Figure 10 The flowchart shown in the embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention illustrates the process of obtaining a multi-scale local feature dataset based on the multi-scale local signal feature map. In step S52, the multi-scale local feature dataset is obtained by processing the multi-scale local signal feature map using the SPP model, including:
[0109] S521. Based on the multi-scale local signal feature map, a multi-scale pooling region is obtained by using spatial pyramid pooling.
[0110] S522. Based on the multi-scale pooling region, obtain the multi-scale local feature dataset through max pooling or average pooling.
[0111] S6. Perform feature fusion on the global feature dataset and the multi-scale local feature dataset to obtain a comprehensive feature vector;
[0112] S7. Based on the comprehensive feature vector, threat events are classified using a multilayer perceptron (MLP) to obtain the prediction results of pipeline threat event types.
[0113] Specifically, such as Figure 11 The figure shows the average accuracy and loss curves of the Transformer-TCN-SPP model in the embodiment of the intelligent identification method for oil and gas pipeline threat events of the present invention. On the training set, a 5-fold cross-validation method was used to evaluate the model and select hyperparameters. In this process, a grid search heuristic algorithm was used to search for the optimal combination of hyperparameters. After completing the 5-fold cross-validation, the results of the five validations were combined, and the model's average accuracy, average recall, average F1-score, and standard deviation were calculated, as shown in Table 2, to obtain an unbiased estimate of the model's performance. The recall and F1-score are relatively close to the accuracy, indicating a balanced model performance. Meanwhile, the small standard deviation indicates that the fluctuations between folds are not significant, and the model is more stable. The cross-validation method is used to determine the optimal hyperparameters of the model.
[0114] Table 2 Cross-validation results of model training
[0115] Cross-validation folds Accuracy Recall F1 score 1 0.994 0.992 0.993 2 0.992 0.991 0.991 3 0.993 0.992 0.993 4 0.995 0.994 0.995 5 0.991 0.990 0.991 Mean 0.993 0.992 0.993 Standard deviation 0.0015 0.0014 0.0015
[0116] With 60 epochs of training, the Transformer-TCN-SPP model exhibits stable performance improvement: the loss curve quickly decreases in the early epochs and then stabilizes, while the accuracy reaches a plateau early and consistently maintains a high level throughout the subsequent epochs. Ultimately, the Transformer-TCN-SPP model training accuracy converges to 99.3%, indicating effective convergence, low loss, and high accuracy.
[0117] Further, after multiple training iterations, the Transformer-TCN-SPP model achieves optimal performance. The Transformer-TCN-SPP model achieves an identification rate of 99.2% on the test set, demonstrating its high reliability in pipeline event classification, similar to the identification rate on the training set, and without overfitting. Table 3 provides detailed evaluation scores, including Precision, Recall, F1 Score, and Support for each event type, ensuring a comprehensive evaluation of the model's performance across different categories. Precision and Recall reflect the model's performance in identifying relevant events (true positives) while minimizing false positives and false negatives. For example, the noise category achieves perfect scores (Precision = 1.000, Recall = 1.000, F1 Score = 1.000), highlighting the model's flawless performance in detecting noise events; other categories, such as cars and machinery, also perform well, with an F1 Score of 0.984, demonstrating a balance between Precision and Recall.
[0118] Table 3 Model Test Results Table
[0119]
[0120] Further, in the model performance comparison phase, we evaluated the performance differences between commonly used time series processing models, including LSTM, Bi-LSTM, and 1D-CNN, and the Transformer-TCN-SPP model. The training and test datasets for the comparison phase were identical and included four event signals: background noise, artificial knocking or digging, car passing, and mechanical digging, with 3000 data samples for each event and a 7:3 dataset division ratio. The model training hyperparameters and environment configurations for the comparison were consistent with the Transformer-TCN-SPP model, as shown in Table 4.
[0121] Table 4 Model Hyperparameters and Environment Table
[0122]
[0123] After several rounds of hyperparameter adjustment, the best training effect is finally achieved. The trained model is tested on the test data set, and the experimental results are shown in Table 5. The comparison of model performance shows that the Transformer-TCN-SPP model is superior to LSTM, Bi-LSTM and 1D-CNN model in event recognition accuracy and test efficiency: the test time of the Transformer-TCN-SPP model is 4.33 milliseconds, and the highest recognition accuracy is 99.2%; in contrast, although the LSTM model is faster than Bi-LSTM, the recognition accuracy is lower, only 90.2%, and the test time is 6.89 milliseconds; the Bi-LSTM model is more accurate (accuracy is 95.6%) than the LSTM, but the test time is the longest, 8.72 milliseconds; the test time of the 1D-CNN model is the shortest (2.27 milliseconds), but its accuracy is low, only 90.7%.
[0124] Table 5 Comparison of model performance
[0125] Model Accuracy Test time Transformer-TCN-SPP 0.992 4.33 ms LSTM 0.902 6.89 ms Bi-LSTM 0.956 8.72 ms 1D-CNN 0.907 2.27 ms
[0126] Therefore, the Transformer-TCN-SPP is obviously superior to LSTM and Bi-LSTM in accuracy and inference time, especially in processing speed and accuracy, and is very suitable for scenarios that require efficient processing of large-scale time series data. This result shows that the new model has made a significant breakthrough in the accuracy and efficiency of threat event classification.
[0127] Further, as shown in the confusion matrix comparison chart in the pipeline event classification in the oil and gas pipeline threat event intelligent identification method embodiment of the present application, Figure 12 based on LSTM, Bi-LSTM, 1D CNN and Transformer-TCN-SPP model, the pipeline event classification results are compared, and the confusion matrix in the pipeline event classification is formed. It can be seen from Figure 12 that the three models of LSTM, Bi-LSTM and 1D CNN have difficulty in distinguishing between vehicle passing and mechanical operation signals; in contrast, the Transformer-TCN-SPP model significantly improves the recognition rate of these two types of events, making it easier to distinguish. The comparison of experimental results shows that the Transformer-TCN-SPP model has obvious advantages in accurately identifying pipeline events, and this comparison further proves the feasibility and superiority of the Transformer-TCN-SPP model in real-time pipeline event detection.
[0128] As shown in Figure 13The system block diagram of the oil and gas pipeline threat event intelligent identification system embodiment of the application is shown, and the application provides an oil and gas pipeline threat event intelligent identification system, which is applied to an oil and gas pipeline threat event intelligent method, and the system comprises a vibration signal acquisition module, a preprocessing module, a global feature extraction module, a short-term signal decomposition module, a multi-scale local feature extraction module, a feature fusion module and a threat event classification module, in particular,
[0129] The vibration signal acquisition module is used for distributing optical fiber sensors on the oil and gas pipeline, adopting optical fiber acoustic sensing technology to obtain the vibration signal of the oil and gas pipeline.
[0130] The preprocessing module is used for denoising and standardizing the oil and gas pipeline vibration signal to obtain a pretreated signal.
[0131] The global feature extraction module is used for extracting global features from the pretreated signal through a Transformer model to obtain a global feature dataset.
[0132] The short-term signal decomposition module is used for decomposing the pretreated signal into a short-term high-energy periodic signal.
[0133] The multi-scale local feature extraction module is used for extracting multi-scale local features from the short-term high-energy periodic signal through a TCN-SPP model to obtain a multi-scale local feature dataset.
[0134] The feature fusion module is used for fusing the global feature dataset and the multi-scale local feature dataset to obtain a comprehensive feature vector.
[0135] The threat event classification module is used for classifying the threat event through a multi-layer perceptron MLP according to the comprehensive feature vector to obtain a pipeline threat event type prediction result.
[0136] The application provides an oil and gas pipeline threat event intelligent identification method and system, which carries out access decision of cross-domain Internet of Things through a cross-domain Internet of Things access decision model based on a space-time enhanced neural network, adopts multi-scale convolution operation and residual connection neural network to obtain space features, adopts BiLSTM to model and obtain time features, adopts a Concat module and a Transformer module to perform feature fusion and enhancement and obtain space-time enhanced features, performs nonlinear mapping of the space-time enhanced features through an output module, and outputs an access decision result, so that the performance of the equipment in processing time series data and the effectiveness of the equipment access decision are improved, and the safety of the equipment is ensured; meanwhile, the model aggregates model parameters of multiple local devices through a personalized federated learning module, effectively balances local optimization and global generalization, and improves the adaptability and cross-domain robustness of the model in the cross-domain Internet of Things scene; the application can process non-independent and identically distributed data, supports cross-domain cooperation under the premise of ensuring data privacy, integrates space, time and space-time feature modeling, and provides an efficient multi-domain cooperation solution.
[0137] It can be understood that the application is described by the above embodiments, which should not be interpreted as limiting the embodiments and the scope of the application. Those skilled in the art know that various changes or equivalent replacements can be made to the features and embodiments without departing from the spirit and scope of the application. In addition, the features and embodiments can be modified to adapt to specific conditions and materials under the guidance of the application without departing from the spirit and scope of the application. Therefore, the application is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the application are within the scope of the application.
Claims
1. A method for intelligent identification of oil and gas pipeline threat events, characterized in that, The method includes: S1. Distribute fiber optic sensors on oil and gas pipelines and use fiber optic acoustic sensing technology to obtain vibration signals of oil and gas pipelines. S2. The vibration signal of the oil and gas pipeline is subjected to noise reduction and standardization preprocessing to obtain a preprocessed signal; S3. Based on the preprocessed signal, extract global features using the Transformer model to obtain a global feature dataset; S4. Perform short-term signal decomposition on the preprocessed signal to obtain a short-term high-energy periodic signal; S5. Based on the short-term high-energy periodic signal, extract multi-scale local features using the TCN-SPP model to obtain a multi-scale local feature dataset; S6. Perform feature fusion on the global feature dataset and the multi-scale local feature dataset to obtain a comprehensive feature vector; S7. Based on the comprehensive feature vector, threat events are classified using a multilayer perceptron (MLP) to obtain the prediction results of pipeline threat event types. S5 includes: S51. Based on the short-term high-energy periodic signal, a multi-layer temporal convolutional network (TCN) model is adopted, and multi-scale local signal feature maps are obtained through processing by the multi-layer temporal convolutional network (TCN). S52. Based on the multi-scale local signal feature map, the multi-scale local feature dataset is obtained by processing it through the SPP model.
2. The intelligent identification method for oil and gas pipeline threat events according to claim 1, characterized in that, In step S2, the vibration signal of the oil and gas pipeline is subjected to noise reduction and standardization preprocessing to obtain a preprocessed signal, including: S21. The vibration signal of the oil and gas pipeline is subjected to noise removal processing to obtain the noise-removed vibration signal of the oil and gas pipeline. S22. Based on the noise-reduced oil and gas pipeline vibration signal, standardize the signal to unify the scale and obtain a preprocessed signal.
3. The intelligent identification method for oil and gas pipeline threat events according to claim 1, characterized in that, In step S4, the preprocessed signal is decomposed into a short-term signal to obtain a short-term high-energy periodic signal, including: S41. Based on the preprocessed signal, divide it according to a predetermined sliding window size to obtain local short-term signal segments; S42. Based on the local short-term signal segment, the energy intensity of the local short-term signal segment is obtained using formula (1). In the formula, E represents the energy intensity, N represents the total number of signal sample points, and x n This represents the signal amplitude at the nth time step; S43. Based on the local short-term signal segment, the periodic characteristics of the local short-term signal segment are obtained through frequency domain analysis; S44. Based on the energy intensity of the local short-term signal segment, the signal-to-noise ratio of the local short-term signal segment is obtained using formula (2). In the formula, SNR represents the signal-to-noise ratio, and E signal E represents the energy intensity of a local short-term signal segment. noise Indicates the noise intensity of a local short-term signal segment; S45. Based on the signal-to-noise ratio of the local short-term signal segment, a signal-to-noise ratio threshold is set to obtain an effective local short-term signal segment; S46. Based on the effective local short-term signal segments and the periodic characteristics of the local short-term signal segments, short-term high-energy periodic signals are obtained through analysis and screening.
4. The intelligent identification method for oil and gas pipeline threat events according to claim 3, characterized in that, In step S43, based on the local short-term signal segment, the periodic characteristics of the local short-term signal segment are identified through frequency domain analysis, including: S431. Based on the local short-term signal segment, perform a Fourier transform using formula (3) to obtain the local short-term signal in the frequency domain. Where F(ω) represents the local short-term signal in the frequency domain, f(t) represents the local short-term signal segment, ω represents the angular frequency, and e -iwt Represents a complex exponential function; S432. Based on the local short-term signal in the frequency domain, periodic behavior is identified by detecting significant peaks, and the frequency domain peak value is output to obtain the periodic characteristics of the local short-term signal segment.
5. The intelligent identification method for oil and gas pipeline threat events according to claim 1, characterized in that, The multi-layer temporal convolutional network (TCN) model employs a specific dilation rate configuration.
6. The intelligent identification method for oil and gas pipeline threat events according to claim 1, characterized in that, In step S51, based on the short-term high-energy periodic signal, a multi-layer temporal convolutional network (TCN) model is used to process the signal, resulting in a multi-scale local signal feature map, including: S511. Based on the short-term high-energy periodic signal, capture the time dependence of the short-term high-energy periodic signal through causal convolution and dilated convolution to obtain the characteristics of the short-term high-energy periodic signal at different time scales. S512. Based on the characteristics of the short-term high-energy periodic signal at different time scales, a multi-scale local signal feature map is obtained by adjusting the expansion factor.
7. The intelligent identification method for oil and gas pipeline threat events according to claim 1, characterized in that, In step S52, based on the multi-scale local signal feature map, a multi-scale local feature dataset is obtained through SPP model processing, including: S521. Based on the multi-scale local signal feature map, a multi-scale pooling region is obtained by using spatial pyramid pooling processing. S522. Based on the multi-scale pooling region, obtain the multi-scale local feature dataset through max pooling or average pooling.
8. An intelligent identification system for oil and gas pipeline threat events, used to implement the intelligent identification method for oil and gas pipeline threat events as described in any one of claims 1-7, characterized in that, The system includes: The vibration signal acquisition module is used to distribute fiber optic sensors on oil and gas pipelines and obtain vibration signals of oil and gas pipelines using fiber optic acoustic sensing technology. The preprocessing module is used to perform noise reduction and standardization preprocessing on the vibration signal of the oil and gas pipeline to obtain a preprocessed signal; The global feature extraction module is used to extract global features based on the preprocessed signal using the Transformer model to obtain a global feature dataset. The short-term signal decomposition module is used to decompose the preprocessed signal into a short-term high-energy periodic signal. The multi-scale local feature extraction module is used to extract multi-scale local features based on the short-term high-energy periodic signal using the TCN-SPP model to obtain a multi-scale local feature dataset. The feature fusion module is used to fuse the global feature dataset and the multi-scale local feature dataset to obtain a comprehensive feature vector. The threat event classification module is used to classify threat events using a multilayer perceptron (MLP) based on the comprehensive feature vector, and obtain the prediction result of pipeline threat event type.