ACO-based feature selection and neural network modeling for accurate gamma-radiation based pipeline monitoring in the oil industry

Abstract

This work presents a novel technique to improve oil pipeline monitoring capabilities, a vital activity in the oil and gas sector. Using Monte Carlo simulations, the work meticulously records data from a pipeline testing environment with various petroleum products and volume ratios. We apply the presented technique to mix four petroleum products—ethylene glycol, gasoline, crude oil, and gasoil—in different volumetric fractions to precisely determine their volume ratios. Many characteristics of the signal, including its mean, standard deviation, autocorrelation, zero-crossing rate, dominant frequency, power spectral density, harmonic-to-noise ratio, cross-frequency coupling, peak-to-peak amplitude, and fall time, are extracted after data collection. To select optimal features, an innovative approach utilizing ant colony optimization is deployed, systematically identifying the most informative feature combinations for volumetric ratio prediction. These meticulously chosen features serve as inputs to a multilayer perceptron (MLP) neural network tasked with accurately determining the volume ratio of the pipeline contents. Remarkably, the methodology showcases remarkable efficacy, with the root mean square error (RMSE) in volume ratio determination found to be less than 0.52. This significant finding not only underscores the robustness of the proposed approach but also promises to revolutionize pipeline monitoring techniques, offering unprecedented accuracy and efficiency in oil industry operations. This research thus represents a pivotal advancement in the field, with far-reaching implications for both academic research and practical applications within the oil and gas sector.