Oil Pipeline Leakage Detection And Location Based On Time-frequency Analysis

As the safe and economic transport tool, pipeline transportation has an obvious advantage in conveying oil and gas. Due to the fast development of crude oil pipeline transportation and the unceasing expansion of pipeline construction, a large and complicated pipeline network is formed in China. However large amount of pipeline leaks caused by wear of long-time running, natural aging of facility, changes of geography and atmosphere and man-made damage have emerged frequently in recent years. An effective approach to avoid economic losses and ensure the safety is the real-time pipeline inspection which can diagnose the leakage faults in time and locate the leakage position exactly. Among the various leak detection and location technologies of oil and gas pipelines, the method based on software had become one of the most widely used methods.The research on crude oil pipeline leakage detection and location applying the technologies of time-frequency analysis and neural network is carried out based on the detailed analysis of site-collected signals. The main endeavors accomplished can be generized as follows:1. The pressure signals are pre-processed based on blind source separation: There is only one observation signal because only one pressure transducer is instilled in upstream and downstream of the pipeline, so it does not satisfy the requirement of blind source separation. In order to settle this puzzle, the Empirical Mode Decomposition is applied to decompose the noised pressure signal into a number of intrinsic mode functions(IMF), and then reconstruct these IMFs into two mixed observation signals. At last the two mixed observation signals is separated into noised signal and real pressure signal by the method of Independent Component Analysis(ICA), therefore the de-noising of pressure signals is come true.2. The pipeline leakage is detected using Hilbert-Huang Transform and Probabilistic Neural Network(PNN): The pipeline operation conditions are classified into four types: leakage state, valve adjusting state, normal state and pump adjusting state. At first the feature vectors of these four type signals are extracted using the technology of Hilbert-Huang Transform. Then the Probabilistic Neural Network is chosen as the classifier whose input is the extracted feature vectors and then is trained by history signals. At last the detecting signal is distinguished as a leakage signal, a normal signal, valve adjusting signal or a pump adjusting signal by well-trained classifier.3. The pipeline leakage position is located using S Transform and Generalized Regression Neural Network(GRNN): The S Transform was applied to obtain the moment when the pressure signal starts to drop caused by leakage. Therefore the time difference between the reception of negative pressure waves by transducers in up and down streams was get. On the basis of the modified negative pressure wave velocity, the leak positioning formula is fitted by Generalized Regression Neural Network whose parameter was optimized by Fruit Flies Optimization Algorithm(FOA). At last a leakage location model based on GRNN-FOA is established,whose input is the time difference calculated by S Transform and the output of which is the leakage position.