Estimación de variables determinantes en la caída de presión asociada al sistema de filtros de una turbina a gas

Abstract

A methodology was developed to approximate, through artificial neural networks to estimate the pressure drop in the system of air filters, from operative data and increase the availability of the electric generation unit. It starts with 28 predictor variables and target variable, based on a sensitivity analysis of the number of predictors is reduced and the design of the structure of the multilayer perceptron concrete. It determines how many and which predictor variables are necessary for a fit with a coefficient of determination more than 0.99 found that one is the minimum amount required, the final structure is 1-12-1 with a coefficient of determination of 0,997 40. The model developed to simulate the variable and provides a powerful tool for increasing the operational availability of generation units in regard to air filtration.