Small leaks in natural gas pipelines are hard to detect, and there are few studies on this problem in the literature. In this paper, a method based on the extended Kalman filter (EKF) is proposed to detect and locate small leaks in natural gas pipelines. First, the method of a characteristic line is used to establish a discrete model of transient pipeline flow. At the same time, according to the basic idea of EKF, a leakage rate is distributed to each segment of the discrete model to obtain a model with virtual multi-point leakage. As such, the virtual leakage rate becomes a component of the state variables in the model. Secondly, system noise and measurement noise are considered, and the optimal hydraulic factors such as leakage rate are estimated using EKF. Finally, by using the idea of an equivalent pipeline, the actual leakage rate is calculated and the location of leakage on the pipeline is assessed. Simulation and experimental results show that this method can consistently predict the leakage rate and location and is sensitive to small leakages in a natural gas pipeline.
Leakage detection in underwater oil and natural gas pipelines using convolutional neural networks
