Oil and gas pipeline safety early warning system and method based on small sample data features

Abstract

The application provides a kind of oil and gas pipeline safety early warning system and method based on small sample data characteristics, relating to oil and gas pipeline warning field, the system includes vibration sensor, station transmission center and terminal control center;Vibration sensor is arranged along the pipeline, for detecting vibration signals in the direction along the pipeline and the vertical direction;Station transmission center is used to obtain the electric signal sent by vibration sensor, and transmit the electric signal to terminal control center;Terminal control center includes central processing unit, for matching the received electric signal with the data in pre-stored abnormal sample database, and judging whether the vibration signal is abnormal according to the preset baseline vibration threshold;A historical data analysis module is also provided in the central processing unit, for small sample data augmentation on abnormal samples, to enrich the abnormal sample database.The application can accurately and efficiently determine the running state of oil and gas pipeline, and detect abnormal events.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of oil and gas pipeline early warning, and particularly relates to an oil and gas pipeline safety early warning system and method based on small sample data characteristics. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] Oil and gas pipelines are an important part of modern industrial systems and play a crucial role in energy transportation and economic development. However, oil and gas pipelines are at high risk of being damaged by third-party intrusion, such as mechanical construction, manual excavation, malicious damage, and heavy vehicle crushing, etc. In addition, natural disasters such as landslides and mudslides also seriously endanger pipeline safety, which may cause oil and gas leakage, environmental pollution, and personal injury, etc. Traditional safety monitoring methods mainly rely on manual patrol, which requires a large amount of manpower and material resources and is time-consuming. When the pipeline is damaged, the line patrol workers may not be able to discover it in time, causing serious damage to the pipeline. Therefore, effective early warning measures need to be taken to ensure the safe operation of oil and gas pipelines.

[0004] Currently, there are some oil and gas pipeline safety early warning methods based on optical fiber sensing, but the optical fiber sensing scheme has problems such as high false alarm rate, serious background noise influence, complex system, high price, and the need for a laser emitter. Once the laser emitter fails, the entire optical fiber early warning system will be paralyzed. Moreover, the optical fiber sensing technology cannot timely monitor the position displacement before landslides and mudslides occur, so a new scheme is needed to overcome the above problems. SUMMARY

[0005] In order to solve the problems of the prior art, the present application provides an oil and gas pipeline safety early warning system and method based on small sample data characteristics, which can realize high-precision early warning monitoring of abnormal events.

[0006] According to some embodiments, the first aspect of the present application provides an oil and gas pipeline safety early warning system based on small sample data characteristics, comprising: a vibration sensor, a station transmission center and a terminal control center; the vibration sensor is arranged along the pipeline and is used to detect vibration signals in the direction along the pipeline and the vertical direction; the station transmission center is used to acquire the electric signal emitted by the vibration sensor and transmit the electric signal to the terminal control center; the terminal control center comprises a central processor, which is used to match the received electric signal with the data in the pre-stored abnormal sample database and judge whether the vibration signal is abnormal according to the pre-set baseline vibration threshold; the central processor further comprises a historical data analysis module, which is used to perform small sample data augmentation on the abnormal samples to enrich the abnormal sample database.

[0007] In a second aspect, the present application provides a method for oil and gas pipeline safety warning based on small sample data characteristics, which is based on the system for oil and gas pipeline safety warning based on small sample data characteristics provided in the first aspect, and comprises:

[0008] Collecting vibration signals along the oil and gas pipeline;

[0009] Extracting vibration characteristic parameters of the vibration signals and performing data dimension reduction;

[0010] Judging whether the vibration signals are abnormal according to a preset baseline vibration threshold, and matching the extracted vibration characteristic parameters with data in the normal sample database and the abnormal sample database, and issuing a warning signal if it is judged that an abnormality occurs.

[0011] Compared with the prior art, the present application has the following advantages:

[0012] The present application provides a system and method for oil and gas pipeline safety warning based on small sample data characteristics, which can accurately and efficiently determine the running state of the pipeline and detect abnormal events by setting a baseline vibration threshold to eliminate the influence of the vibration of the pipe body itself and environmental factors, and using a small sample data expansion method combined with a machine learning algorithm. The present application has the following advantages: (1) wider application range, the sensor is arranged alone around the pipe body, and is independent of the pipe material and the internal medium, and is not affected by the pipe material and the medium; (2) better real-time performance, which can collect, transmit and process data in real time, and can monitor the vibration state of the surrounding environment of the pipeline in real time; (3) can monitor the displacement before the occurrence of geological disasters such as landslides and debris flows, and timely issue a warning signal to ensure the safety of the pipeline; (4) the vibration signals collected in real time are extracted and input into a trained classification warning model to identify the type of intrusion event; (5) through the historical data analysis module, small sample expansion processing is performed on the occasional abnormal events with small sample size, and the algorithm model is continuously optimized and adjusted to improve the accuracy and stability of the feature model.

[0013] The advantages of the additional aspects of the present application will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0014] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the description of the exemplary embodiments of the present application and their description serve to explain the present application, and do not constitute an improper limitation of the present application.

[0015] Figure 1 It is a whole schematic diagram of the system in Example 1;

[0016] Figure 2This is a schematic diagram of the signal processing process;

[0017] Figure 3 This is a flowchart of the method in Example 2;

[0018] Figure 4(a) is a schematic diagram of the signal features extracted from manual mining.

[0019] Figure 4(b) is a schematic diagram of the extracted features of the mechanical excavation signal.

[0020] The system includes: 1. Pipeline; 2. Vibration sensor; 3. Station transmission center; 4. Data acquisition card; 5. Station storage system; 6. Signal transmission system; 7. Central processing unit; 8. Terminal control center; and 9. Real-time monitoring system. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Example 1

[0023] like Figures 1-2 As shown in Embodiment 1 of the present invention, an oil and gas pipeline safety early warning system based on small sample data features is provided, including: a vibration sensor 2, a station transmission center 3, and a terminal control center 8; the vibration sensor 2 is installed along the pipeline 1 to detect vibration signals along the pipeline direction and vertically; the station transmission center 3 is equipped with a data acquisition card 4 connected to the vibration sensor 2 to acquire the electrical signals emitted by the vibration sensor, and transmits the electrical signals to the terminal control center 8 through a signal transmission system 6 installed in the station transmission center 3; the terminal control center 8 includes a central processing unit 7, which is used to match the received electrical signals with data in a pre-stored abnormal sample database, and determine whether there is an abnormality in the vibration signal according to a preset baseline vibration threshold. If an abnormality is determined to occur, an early warning signal is issued.

[0024] The vibration sensors 2 are uniformly arranged along the pipeline 1 in the same trench. Each vibration sensor 2 can detect the vibration signals in the pipeline direction and the vertical direction and convert them into corresponding electrical signals. The sensors are connected to the data acquisition cards in the transmission center of each station through wiring. The vibration sensors use high-sensitivity and low-distortion sensors, such as piezoelectric sensors, magnetic sensors, acceleration sensors, etc. The sensors have corrosion resistance, high temperature resistance, and anti-interference characteristics. In this embodiment, IEPE acceleration sensors are selected. Multiple vibration sensors are arranged at different positions to collect vibration signals. The spacing of each vibration sensor is determined according to the sensitivity of the selected sensor and the soil medium density, moisture content, and other indicators around the pipeline to ensure the stability and reliability of the collected signals. The signal range that can be collected by the sensor in different soil media is different. The specific arrangement method needs to be determined according to the actual soil medium density, moisture content, and other indicators. In the case of hard soil, the signal attenuation is low, and the spacing can be appropriately increased. Correspondingly, the distance between the sensors can be appropriately reduced in high-risk areas where risks frequently occur. In actual arrangement, the spacing needs to be appropriately increased or decreased based on experience. For areas prone to mudslides and landslides, sensors with GPS function are used. In the potential landslide area, one or more dual-frequency GPS receivers are installed on the vibration sensor 2 to measure and record the position information of the ground surface regularly. The dual-frequency GPS receiver continuously measures the position of the ground surface, and the position data are transmitted to the terminal control center together with the vibration data. The central processor in the terminal control center processes and analyzes the collected data, including differential processing, data filtering, and removing multi-path interference, to obtain accurate ground surface displacement information. Using the signals of multiple satellites, the three-dimensional displacement of the ground surface is calculated, including the east-west, north-south, and vertical displacements. The continuous displacement data are analyzed in time sequence to detect the change trend of the ground surface displacement and identify potential landslide signs.

[0025] A signal conditioner with high precision, low noise, and high sampling rate is installed on each vibration sensor, and is connected to the central processor through the data acquisition card. The data acquisition card selects a card with high sampling rate, low noise, and high precision, such as a PCI or PCIe interface data acquisition card. At the same time, a card suitable for the type and range of sensor output signals is selected, such as a card with an analog signal amplifier and filter to meet the level and frequency requirements of the sensor output signal. The sampling rate of the data acquisition card should be higher than the output sampling rate of the signal conditioner to ensure the quality and accuracy of the collected signals. Each data acquisition card can simultaneously connect 16-32 sensors in the forward and backward directions along the pipeline laying direction.

[0026] The station transmission center includes various types of process stations established along the oil and gas pipeline, such as monitoring stations, water injection stations, pigging stations, heating stations, and metering stations. The station transmission center 3 is equipped with a station storage system 5. Data acquisition cards 4 located at each station transmission center collect electrical signals and store them in the station storage system. Simultaneously, the signals are transmitted to the central processing unit 7 of the terminal control center 8 via a signal transmission system 6 for analysis and processing.

[0027] Each station's transmission center can deploy multiple data acquisition cards simultaneously to ensure that it can receive signals collected by all sensors along the pipeline at the same time, achieving comprehensive monitoring of the entire pipeline without blind spots, while ensuring the accuracy and stability of the data.

[0028] The terminal control center 8 is equipped with a real-time monitoring system 7 and a central processing unit 9. The real-time monitoring system 7 receives real-time signals sent by the transmission centers 3 at each station and transmits them to the central processing unit 9 for analysis and processing. The central processing unit is equipped with an algorithm model and an early warning system. The algorithm model is used to process and analyze the vibration signal data to determine whether the pipeline has been damaged. If an abnormality is detected, an early warning signal is issued through the early warning system.

[0029] The collected signals are trained using machine learning algorithms to establish normal and abnormal sample databases. Real-time comparative analysis of the collected signals is performed, and anomalies are determined based on a preset baseline vibration threshold. Once an abnormal signal is detected, it is matched against event categories in the abnormal sample database (such as third-party intrusion events, landslides, and debris flows). The central processing unit automatically issues an early warning signal, transmitting detailed information such as the specific abnormal event and its location to relevant personnel. Simultaneously, real-time pressure and temperature data from the pipeline's operation are analyzed to determine if there are pressure and temperature changes associated with unusual vibration patterns, thus improving the accuracy of the early warning. In actual pipeline operation, the temperature and pressure inside the pipe typically maintain a relatively stable trend. When the pipeline is compromised, sensors deployed around the pipe will detect vibration signals. Depending on the type of damage, the pressure and temperature inside the pipe will exhibit different trends; for example, when the pipe is compromised and leaks, the pressure and temperature will decrease. Therefore, abnormal vibration signals are combined with pressure and temperature changes to improve the accuracy of the early warning. Early warning signals can be delivered to operators and other relevant personnel using various methods, including visual, auditory, and vibration signals.

[0030] The central processing unit also includes a historical data analysis module, used to augment abnormal samples with small sample data to enrich the abnormal sample database. Through long-term monitoring and analysis of vibration data, small sample data augmentation is performed on abnormal vibration signals to increase the number of abnormal vibration signal samples, thereby improving the system's accuracy and robustness.

[0031] The early warning system includes an audible and visual alarm device and a communication module. The audible and visual alarm device can emit a clear sound and flashing light to alert operators; the communication module can send alarm information to the terminal control center or other designated station transmission centers.

[0032] Example 2

[0033] This embodiment provides a method for early warning of oil and gas pipeline safety based on small sample data features. This method is based on an oil and gas pipeline safety early warning system based on small sample data features provided in Embodiment 1, and includes the following steps:

[0034] S1. Collect vibration signals along the oil and gas pipeline;

[0035] S2. Extract the vibration characteristic parameters of the vibration signal and perform data dimensionality reduction;

[0036] S3. Determine whether there is an abnormality in the vibration signal based on the preset baseline vibration threshold, and match the extracted vibration feature parameters with the data in the pre-existing normal sample database and abnormal sample database. If an abnormality is detected, issue an early warning signal.

[0037] In step S1, after collecting vibration signals along the oil and gas pipeline, the vibration signals are further filtered and denoised to remove interference signals from the environment and the pipeline operation itself. Filtering and denoising methods include mean filtering, median filtering, wavelet denoising, Kalman filtering, low-pass filtering, high-pass filtering, and band-pass filtering. The specific filtering and denoising method can be selected based on the output signal type of the selected sensor and the environment. For example, for the IEPE accelerometer used in Example 1, its data has no baseline drift; therefore, a band-stop filter is directly used for filtering, and wavelet packets are used for denoising to eliminate signal noise and improve the signal-to-noise ratio.

[0038] In step S2, time-domain, frequency-domain analysis, and empirical mode analysis methods are used to process the vibration signals output by each vibration sensor for the filtered signals, and vibration characteristic parameters such as vibration amplitude, vibration frequency, phase, and energy are extracted. For different third-party intrusion conditions, the amplitudes and frequencies of the vibration signals are different. Therefore, by extracting characteristic parameters such as the amplitude, frequency, phase, and energy of the vibration signals, it is possible to effectively identify whether the occurring vibration belongs to an abnormal condition. For the signals output by sensors in areas prone to debris flows, landslides, etc., in addition to the above-mentioned feature extraction, GPS position feature extraction is also performed. The collected GPS data is processed and analyzed, including differential processing, data filtering, removal of multipath interference, etc., to obtain accurate surface displacement information. The signals of multiple satellites are used to calculate the three-dimensional displacement of the surface, including east-west, north-south, and vertical displacements. The continuous three-dimensional displacement data is subjected to time-series analysis to detect the change trend of the surface displacement and identify potential landslide signs.

[0039] Data dimensionality reduction processing is performed on the extracted features to reduce the computational amount. Dimensionality reduction processing is performed on the data of time-domain, frequency-domain, and empirical mode analysis. Reducing the feature dimension can reduce the computational amount. Especially in processing large-scale data sets or real-time monitoring applications, dimensionality reduction can save resources, and reducing the computational burden can improve the processing speed, making data processing and storage more efficient. In the process of dimensionality reduction processing, each sample in the data set obtained after processing the data is represented as a high-dimensional feature vector. Subsequently, the data is standardized, and appropriate dimensionality reduction methods are selected, mainly including: principal component analysis (PCA), independent component analysis (ICA), t-SNE, UMAP, etc.; Subsequently, the dimensionality reduction method will calculate a mapping matrix to map the high-dimensional features to a low-dimensional space. Before dimensionality reduction: The original data set contains n samples, and each sample has m-dimensional features. The original data can be represented as an n×m matrix, where each row is a sample and each column is a feature. After dimensionality reduction: After dimensionality reduction, the feature dimension of each sample is reduced to p (p < m), and an n×p low-dimensional data matrix is obtained. Each row still represents a sample, but each column now represents a low-dimensional feature.

[0040] In step S3, first, based on the historical data analysis of the pipeline and the application of machine learning algorithms, a baseline vibration threshold is set, which can be determined according to multiple factors such as the working state and environment of the pipeline.

[0041] Based on a preset baseline vibration threshold, the system determines whether the acquired vibration signals are abnormal. Machine learning algorithms are then used to extract, classify, and discriminate features from the processed data, distinguishing between normal and abnormal signals. During normal pipeline operation, the medium transported within the pipeline also causes vibrations in the surrounding environment. Therefore, by setting a baseline vibration threshold, vibrations caused by the pipeline itself and the natural environment are eliminated, reducing the amount of data processing. The machine learning algorithm learns from a large number of operational data samples, identifies the occurrence of abnormal events, and categorizes non-abnormal data within the baseline vibration threshold. Machine learning algorithms that can use include Support Vector Machine (SVM), Decision Tree, Random Forest, Neural Networks, K-Nearest Neighbor (KNN), Naive Bayes, Ensemble Learning, and Clustering. In application, the appropriate machine learning algorithm is selected based on the actual data characteristics, and reasonable parameter adjustments and model selection are made to achieve the best prediction results.

[0042] Methods for obtaining the baseline vibration threshold include:

[0043] S301. Obtain historical vibration signals;

[0044] S302. Extract the vibration characteristic parameters of the historical vibration signal and perform data dimensionality reduction;

[0045] S303. Cluster the dimensionality-reduced data to determine normal vibration characteristics and abnormal vibration characteristics, and establish a normal sample database and an abnormal sample database.

[0046] S304. Perform small sample data augmentation on the abnormal samples in the abnormal sample database to enrich the abnormal sample database;

[0047] S305. Set a baseline vibration threshold, perform feature learning on features in the normal sample database and the abnormal sample database, and optimize the baseline vibration threshold based on the feature learning results.

[0048] In step S302, the vibration characteristic parameters include time-domain parameters, frequency-domain parameters, and statistical parameters, specifically including: peak frequency, spectral peak, spectral width, mean, variance, standard deviation, etc.

[0049] In step S303, the K-means clustering algorithm is used to cluster the dimensionality-reduced data to determine the normal and abnormal vibration characteristics of the pipeline and the corresponding GPS location characteristics. Based on the obtained normal and abnormal vibration characteristics, a normal sample database and an abnormal sample database are established. The normal sample database stores vibration data of the pipeline during normal operation, while the abnormal sample database stores vibration data generated when the pipeline is invaded by a third party, such as vibrations from mechanical excavation, mechanical picks, directional drilling, and oil theft through boreholes.

[0050] In step S304, the confirmed abnormal vibration signals are augmented using methods such as adding Gaussian white noise, rotation, translation, flipping, and cross-domain migration data augmentation of multimodal vibration data to generate more abnormal vibration signals. This enriches the abnormal sample database, distinguishes event categories (such as third-party intrusion, landslides, and debris flows), and improves the generalization ability and robustness of the identification model. The machine learning algorithm deployed in the central processing unit continuously optimizes and adjusts itself by reading the abnormal sample database, increasing the training load on the model and improving the accuracy and stability of the system.

[0051] In view of the wide geographical area spanned by oil and gas pipeline networks, this embodiment also proposes cross-domain migration of multimodal vibration data, the specific steps of which include:

[0052] Collect vibration data from different oil and gas pipelines, different regions, or different operating conditions. These data may include different types of pipelines, different media, different pipeline materials, etc.

[0053] Feature extraction is performed on vibration data from different sources, followed by standardization and preprocessing to ensure compatibility between different data sources;

[0054] This approach maps different source data to a shared feature space, thereby establishing connections between different data domains. It employs feature selection and dimensionality reduction methods to select the most representative feature subset, and then maps data from different sources to the feature space. It uses a transfer learning model to learn the connections between different data sources by sharing parameters and features, thus mapping different source data to a shared feature space.

[0055] Cross-domain transfer methods are used to generate samples in one data domain that are similar to but different from those in another data domain, thereby expanding the dataset.

[0056] This cross-domain migration data augmentation method for multimodal vibration data can help models better adapt to different vibration conditions in oil and gas pipelines, thereby improving the accuracy and reliability of safety early warning.

[0057] In step S305, machine learning algorithms are used to learn features from both the normal and abnormal sample databases. A historical data analysis module is set up in the machine learning algorithm model deployed on the central processing unit to continuously optimize the threshold by learning and analyzing historical operating data. Since the amount of data generated during the daily operation of oil and gas pipelines is relatively small, the model achieves self-optimization by continuously learning from historical data.

[0058] As shown in Figure 4, this embodiment simulates the signal characteristics generated by two different event types: mechanical excavation and manual excavation. These include time-domain, frequency-domain, and time-frequency-domain signals. From top to bottom in the figure, these represent the time-domain signal, frequency-domain signal, and time-frequency-domain signal, respectively. The time-domain signal is acquired by IEPE accelerometers buried along the pipeline at fixed intervals, filtered by a band-stop filter, and denoised using wavelet packets. The frequency-domain signal is obtained by FFT transformation of the time-domain signal. The time-frequency-domain signal is obtained by STFT transformation, decomposing the signal into frequency components of different time periods. After applying vibration characteristic parameters to the signal, the signal characteristics of different event types show significant differences. A neural network learning method is used to learn the features of the processed signal characteristics, enabling accurate identification of normal events and abnormal intrusion events. In this embodiment, by continuously training the algorithm model, it can identify different event types, improving the accuracy and stability of the alarm system.

[0059] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for early warning of oil and gas pipeline safety based on small sample data features, characterized in that, include: Collect vibration signals along oil and gas pipelines; The vibration characteristic parameters of the vibration signal are extracted, and the data dimensionality is reduced. The vibration signal is judged to be abnormal based on the preset baseline vibration threshold, and the extracted vibration feature parameters are matched with the data in the pre-existing normal sample database and abnormal sample database. If an abnormality is judged to have occurred, an early warning signal is issued. The method for obtaining the baseline vibration threshold includes: Acquire historical vibration signals; The vibration characteristic parameters of the historical vibration signals are extracted, and data dimensionality reduction is performed. Cluster the dimensionality-reduced data to determine normal and abnormal vibration characteristics, and establish normal and abnormal sample databases. To enrich the abnormal sample database, small sample data are used to augment the abnormal samples in the database. Set a baseline vibration threshold, perform feature learning on features in the normal sample database and the abnormal sample database, and optimize the baseline vibration threshold based on the feature learning results; A safety early warning system for oil and gas pipelines based on small sample data features includes: vibration sensors, a station transmission center, and a terminal control center. The vibration sensors are installed along the pipeline to detect vibration signals along the pipeline direction and vertically. The station transmission center acquires the electrical signals emitted by the vibration sensors and transmits them to the terminal control center. The terminal control center includes a central processing unit (CPU) that matches the received electrical signals with data in a pre-stored abnormal sample database and determines whether the vibration signals are abnormal based on a preset baseline vibration threshold. When determining abnormal signals, the CPU also uses real-time pressure and temperature data from the pipeline's operation. The CPU also includes a historical data analysis module for amplifying abnormal samples using small sample data. This amplification involves adding Gaussian white noise, rotation, translation, flipping, and cross-domain migration data amplification of multimodal vibration data to generate more abnormal vibration signals, thus enriching the abnormal sample database.

2. The method for early warning of oil and gas pipeline safety based on small sample data features as described in claim 1, characterized in that, The vibration sensor is equipped with a dual-frequency GPS receiver.

3. The method for early warning of oil and gas pipeline safety based on small sample data features as described in claim 1, characterized in that, The steps for cross-domain migration of the multimodal vibration data include: Collect vibration data from different oil and gas pipelines, different regions, or different operating conditions; Feature extraction is performed on vibration data from different sources, followed by standardization and preprocessing. Map data from different sources to a shared feature space; Generate samples in one data domain that are similar to those in another data domain.

4. The method for early warning of oil and gas pipeline safety based on small sample data features as described in claim 1, characterized in that, It also includes collecting GPS data and using the collected GPS data to calculate the three-dimensional displacement of the Earth's surface.

5. The method for early warning of oil and gas pipeline safety based on small sample data features as described in claim 4, characterized in that, By performing time-series analysis on continuous three-dimensional displacement data, the changing trend of surface displacement can be detected.

6. The method for early warning of oil and gas pipeline safety based on small sample data features as described in claim 1, characterized in that, Machine learning algorithms are used to learn features from normal and abnormal sample databases to optimize the baseline vibration threshold.

Images