Research On Safety Operation Method Of Pipeline Based On Multi-sensor Information Fusion

With the advantages of low cost, energy saving, high safety performance, stable supply, etc, pipeline transportation has become one of the main modes for oil transportation. It has been playing an increasingly important role in the national economy development. However, due to inevitable aging, erosion, damage and bunkering, pipeline leakage occurs frequently.In order to make full use of the feature information, improve the diagnostic accuracy of leakage, and solve the uncertain problem of the pipeline running states, a pipeline system has been designed based on multi-sensor information fusion, and a leak detection method has been proposed based on the D-S evidence theory. Besides, a leakage flow estimation method based on KPCA-SVR has been proposed, aiming at the pipelines in the state of leaking. The main work of this thesis is as follows:According to the comparison of several different leak detection methods, a real-time pipeline system has been designed based on multi-sensor information fusion, which is characterized by hierarchical information fusion and online-offline coordination.A feature extraction method has been proposed to characterize the comprehensive information of the pipeline signals. The energy feature of different frequency bands based on wavelet packet decomposition was used to represent the abrupt information of the pipeline signals, while the slope feature based on the least-square polynomial fitting was used to represent the trend information of the pipeline signals. Because of high dimensions and strong nonlinearity, KPCA has been used to reduce the dimensions of the energy feature, which successfully reduces the computation complexity of the leak detection.A leak detection method has been proposed based on support vector machine D-S evidence theory for the real-time pipeline system. This method uses pressure feature and flow feature to build classifiers of unbalanced binary tree SVM. The outputs of the classifiers are the probabilities. In addition, D-S evidence theory is used to fuse the probability outputs in order to get the final judgments of the running states.Each unbalanced binary tree SVM classifier contains several classifiers. Due to object conditions, the sample sizes of different classes are much different from each other, which cause the movement of the classification boundary of SVM to the class with smaller size. Aiming at solving this problem, kernel K-means clustering has been used to process the class with larger size, improving the effectiveness and accuracy of the proposed classification. In terms of parameter optimization of SVM, a method has been proposed based on GA kernel K-means clustering.Accurate judgment of the leakage flow in case of the pipeline leaking will make a lot of senses in incident controlling and formulation of the emergency measures. A leakage flow estimation method based on KPCA-SVR has been proposed, which uses KPCA for the dimension reduction of the pipeline features and SVR for further regression. This method can efficiently reduce the computation complexity of solving model and improve the accuracy of the estimates of leakage flow.