PSO Trained Hybrid Intelligent Classifier Using Wavelet and Statistical Features for Pipeline Leak Classification

Abstract

One of the main problems in the oil industrial field is the leakage intransporting pipelines due to its effect on human society , environment, and money loss.Therefore, the bottleneck for most researches in this subject is to minimize false alarm rate(FAR) for the adopted leak detection method. Although some recent methods succeed inclassifying the existence or absence of the leak as a binary classification problem. But thispaper proposed a novel leak detection technique which predicts the leak location andestimates its size within certain pre-defined ranges. In order to simulate the environmentalconditions for real-time operating oil pipeline, accurate simulator known as OLGAprogram creates the oil physical parameters. Various methods for features extraction areconsidered such as statistical and wavelet techniques which are implemented to get thefeatures from the fluid simulated waveforms. These features are organized and fed to anANN classifier trained by PSO algorithm to determine the leak class out of 10 suggestedclasses. The proposed leak detection technique is used to simulate 18 kilometers belongingto Iraqi crude oil pipelines company operated in Baghdad. The achieved results of the truepositive rate (TPR) for the proposed applied method for the leak detection andclassification of different leak classes in terms of their positions and magnitudes wereabout 97%.