Pipeline Leak Diagnostic Based On Ensemble Unscented Kalman Filtering

Currently pipeline transportation has incomparable natural advantage in the area of numerous ways of oil and gas transportation in oil and gas industry. The industry of pipeline construction has developed rapidly since 1970 s in China. A variety of typical oil and gas transportation pattern has been largely achieved in recent days and a number of regional pipeline network systems have also been built. The industry of pipeline network has been shifted from sparsely populated regions to densely populated areas and from economic less-developed areas to developed areas. The safety issues of pipeline will affect more social and economic benefits. There are many security risks in old pipelines as the construction condition is limited when they were constructed. And new pipelines were mostly constructed in the western region in which the occurrence of natural disasters is more frequent. Besides this, man-made sabotage makes the pipeline leak accidents more serious. Therefore, it has more practical significance of increasing the timeliness and accuracy of leak diagnostic techniques in oil and gas pipelines.This paper does depth research in nonlinear filtering pipeline leak diagnostic techniques based on transient flow equations. The main work is as follows:1. The pipeline state-space model of transient flow is established. A simulation of the case that there is a leak in the pipeline is conducted. This simulation verifies the analysis in the process of modeling and support the foundation for follow-up study. Moreover, the key technologies of leak diagnosis based on nonlinear filtering are also analyzed.2. This paper proposes diagnosing pipeline leak based on SR-UKF. In each filter iteration in UKF, the calculating of the square root of the state error covariance matrix is needed when sampling sigma-points. However, the state error covariance matrix is propagated in the whole UKF. Besides, there exists non-positive definite covariance in the transfer process because of the nonlinear state-space itself. So this paper uses SR-UKF algorithm for improvement. Simulation results show that this method can improve the numerical stability and make sure non-negative definite characteristics of the error covariance matrix compared to UKF.3. This paper proposes diagnosing pipeline leak based on EnKF. When the pipeline is divided to many parts, it needs a large number of samples which leads to long filtering cycle and large storage space. The assimilation method is essentially the same as the best estimation. As a way to reduce the computational cost of the sequential assimilation(SA), ensemble Kalman filter has shown its great potential for business applications. Ensemble Kalman filter is a Monte Carlo implementation of Kalman filter essentially. It can improve computational efficiency and timeliness of the leak diagnosis, and save computational cost.4. This paper proposes diagnosing pipeline leak based on EnUKF. As the ensemble Kalman filter is batch processing for data and it transforms a ‘nonlinear filtering’ problem into a problem of ‘estimation of the statistical properties’. This is similar to unscented transform. By combining them, the estimation of both mean and covariance matrix can be more accurate. This makes positive contributions to both the timeliness and accuracy of pipeline leak detection.