GTFE-Net-BiLSTM-AM: An intelligent feature recognition method for natural gas pipelines

The recognition of pipeline features contributes to its safe management by preventing severe consequences such as leakage resulting from bending deformation and denting under external pressure. However, extracting features of such a facility is complex and challenging when machine learning techniques are applied to feature recognition. Hence, this paper proposes a feature recognition technique for gas pipelines based on Gramian Time Frequency Enhancement Net (GTFE-Net), Bi-directional Long Short-Term Memory (BiLSTM) and attention mechanism (AM), namely GTFE-Net-BiLSTM-AM. Specifically, GTFE-Net is applied to enhance the time-frequency input bending strain signal, which is subsequently incorporated with the BiLSTM model to extract spatio-temporal features. The attention mechanism computes the corresponding weight of output features. The results show that the proposed method's recognition accuracy reaches 93.7%. The comparison study with the existing models validates the proposed method's superiority and shows that its accuracy is higher than that of the existing models (more than 0.9%) or their combined models (more than 1.1%). Overall, the proposed method contributes to the safety, reliability, and operation of natural gas pipelines.