Perimeter intrusion event identification of oil and gas pipelines under complex conditions based on deep transfer learning

The operating conditions along long-distance oil and gas pipeline are completed, and the premise that the distribution of actual samples is consistent with that of standard samples in traditional method is destroyed. This situation results in the low identification accuracy of intrusion event for single identification model under different conditions. In order to improve the identification model deviation, this paper proposes a pipeline intrusion event identification method based on the deep transfer learning for domain invariant feature. The stacked sparse auto-encoder network is utilized to adaptively extract the domain-invariant features for the intrusion events under different working conditions. Then, the transfer learning is introduced to achieve the accurate identification of pipeline intrusion events under complex conditions. The proposed method reduces the distribution difference between complex real scenes and typical scenes through scene difference evaluation, and obtains an effective domain invariant model. The experiment results show that the proposed method can obviously improve the recognition results of oil and gas pipeline intrusion events under complex conditions, and enhance the identification accuracy. © 2019, Science Press. All right reserved.