Leakage diagnosis of natural gas pipeline based on multi-source heterogeneous information fusion

Abstract

Due to long-term service, natural gas pipelines are prone to corrosion, resulting in pipeline leakage failure and environmental pollution. However, it is challenging to provide an accurate leakage diagnosis for existing single-sensor detection techniques. In this paper, we propose a multi-source heterogeneous information fusion method for the complementary fusion of laser optical sensing and weak magnetic technologies. Firstly, the laser and weak magnetic signals are converted into two-dimensional images using continuous wavelet transform (CWT) and then fused in data-level. Secondly, deep reinforcement learning (DRL) combines the perception ability of deep learning and the decision-making ability of reinforcement learning. Consequently, the deep Q-network (DQN) method is proposed as a novel method for leakage diagnosis of natural gas pipelines. Then, an improved capsule network based on dense block is designed for feature enhancement. Finally, experimental results verify the effectiveness of the proposed method in recognizing the formed leakage and potential leakage. Moreover, the results demonstrate that the proposed method outperforms single-sensor-based and state-of-the-art methods in terms of diagnostic accuracy and cross-domain transfer tasks. This will provide a theoretical basis for pipeline leakage failure prevention and maintenance decision-making.