چكيده به لاتين
Abstract:
Massive pipeline networks are currently used around the globe to transfer and distribute natural gas, crude oil and other lightweight petroleum products. Leakage is one of the most important issues affecting the integrity of pipelines. The occurrence of leakage entails numerous hazards including environmental pollution, capital loss and human casualty.
As a result, pipeline monitoring and leak detection are of utmost significance. This has led to the development of various methods of implementing leak detection systems each of which has its own advantages and disadvantages. In this research, the different ways of detecting leaks are introduced and studied. After a survey of the existing methods, a software method based on measuring the pipeline parameters is designed using neural networks. In this method, three distinct neural networks are used for leak detection, leak localization and leak rate estimation. The data used here were obtained from a simulation of the Western Karun-Omidieh pipeline in the OLGA Pipeline Simulation software. The influence of noise on the input data and the performance of each of the neural networks is investigated. The first neural network is tasked with identifying normal and leakage conditions. This network is capable of correctly recognizing 100% and 96.6% of the predefined patterns in the absence and presence of noise, respectively. The second neural network is responsible for localizing the leakage and is able to recognize, in the absence and presence of noise, respectively 100% and 96% of the predefined location patterns. The third neural network is designed to estimate the leakage flow rate and can estimate leakage flow rates as small as 1% of the entry flow. With a mean squared error of 0.58, the leakage flow estimation has an average accuracy of 1.31% at various points along the pipeline.
Keywords: leak detection, pattern recognition, leak localization, pipeline leakage, neural network
