A Multiphase Liquid-Gas Plant Modelling Using Fuzzy Cognitive Maps: An Application to an Actual Experimental Plant

Although the manufacturing sector now reaps the most benefits from digitization, the oil & gas sector is increasingly embracing digital technology to boost system efficiency, particularly when it comes to modeling and simulation. The oil & gas industry is a complex and multiscale system, making it more challenging to construct a complete and accurate model. This paper presents an algorithm based on the combined use of Fuzzy Cognitive Maps (FCMs) and Gray Wolf Optimization (GWO) to identify the minimal causal model for estimating the level and pressure of a vertical tank in a multiphase liquid-gas plant. Two FCMs were modelled to regress tank level and pressure separately, to analyze the minimal causal relationships among the involved variables. By choosing only simulations concerning the most usual working conditions for the plant as the training dataset, an average accuracy in the training phase of about 85% (with peaks of 99%), and 90% in the testing phase, could be achieved.