Ensuring the integrity and reliability of oil and gas pipelines is of paramount importance in safeguarding energy security and protecting the environment. However, these pipelines are often exposed to various threats, such as human sabotage and third-party construction excavations, which may cause severe fires and explosion accidents. Therefore, it is necessary to develop an effective method to detect and identify different types of events occurring along pipelines. In this study, distributed optical fiber vibration monitoring technology is used to collect the waveform signals of six types of events that occur along a 28.9-kilometer-long pipeline. Subsequently, the Gramian angular field (GAF) transform is used to convert the one-dimensional time-series signals into two-dimensional image information, enabling the capture of characteristic patterns for each event. Next, the GoogLeNet deep learning model is used to classify and identify image information and evaluate the recognition accuracy and false alarm rate of the model. This study proposes an efficient and accurate method for oil and gas pipeline threat identification based on the GAF transform and deep learning.
This study proposes a novel method for oil and gas pipeline threat identification based on distributed optical fiber vibration monitoring technology and deep learning. The waveform signals of six types of events (manual excavation, machine damage, noise, walking, vehicle damage, and water flow vibration) were collected along a 28.9-kilometer-long pipeline, which have the potential to jeopardize pipeline safety. To enhance the feature representation of the signals, a filter was employed to remove noise, and then the GAF algorithm was used to convert the one-dimensional time-series signals into two-dimensional images, enabling the capture of the characteristic patterns of each event. Subsequently, three different deep learning networks, GoogLeNet, VGG, and AlexNet, were employed to classify and identify the images and compare their recognition accuracies and false alarm rates. Experiments were conducted to evaluate the performance of our method, demonstrating that GoogLeNet outperformed the other two networks in terms of recognition accuracy and detecting false alarm rates. The effect of the signal-to-noise ratio (SNR) on the classification performance was analyzed. The GoogLeNet network was determined to achieve optimal classification performance when the SNR was 8 dB.
The main contribution of this study is the proposal of a novel method for oil and gas pipeline threat identification based on the GAF algorithm and deep learning. The GAF algorithm was used to transform the waveform signals of six types of field-collected events (manual excavation, machine damage, noise, walking, vehicle vibration, and water flow vibration) into two-dimensional images that captured the characteristic patterns of each event. Subsequently, three different deep-learning networks, GoogLeNet, VGG, and AlexNet, were used to classify and identify the images. Experiments were conducted to evaluate the performance of our method and compare it with existing methods. The experimental results demonstrate that our method has several advantages over existing methods. First, as shown in Table 2, GoogLeNet can achieve a high classification accuracy and recall rate on both the training and test datasets, indicating that our method generalizes well to new data. Second, as shown in Fig. 7, GoogLeNet performs best in terms of accuracy and loss in the training process, indicating that the GoogLeNet model has strong fitting and generalization abilities. Third, as shown in Figs. 8?13, GoogLeNet can obtain high AUC values for machine damage and manual excavation events, can completely and accurately identify manual excavation events, and has a low false-positive rate (2.78%) for machine damage events, which are considered the most critical threats to pipeline safety. This indicates that the proposed method can effectively distinguish between different types of events. Fourth, as shown in Fig. 14, GoogLeNet can achieve optimal classification performance when the SNR is 8 dB, indicating that our method can handle noisy signals well.
In conclusion, this study uses a distributed optical fiber vibration sensing system to collect six types of event signals along a 28.9-kilometer-long natural gas pipeline and transforms them into two-dimensional image information through the GAF transform. It then uses three deep learning models, GoogLeNet, VGG, and AlexNet, to classify and identify the image information and compare the recognition accuracy and false alarm rate of each model. The results show that the GoogLeNet model outperforms the other two models in terms of recognition accuracy and false alarm rate, achieving a recognition accuracy of 97.79% and effectively differentiating between manual excavation and machine excavation events when the system signal-to-noise ratio is greater than 8 dB. It is suitable for the real-time identification of threat events along the pipeline. The method proposed in this paper provides a novel conceptual framework and technical approach for the safety monitoring of long-distance pipelines.
