Model predictive control to ensure high quality hydrogen production for fuel cells

In this work, a conventional plant wide control of a hydrogen production process from bioethanol is analyzed. The objective is to determine if the carbon monoxide (CO), in the produced hydrogen, exceeds the Proton Exchange Membrane Fuel Cell quality requirement of 10 ppm. Commercial sensors that meet those process conditions at high temperature are not easily available. Then, the development of two soft sensors, based on neural network, for online estimation of CO concentration in the H2 stream is presented. Higher CO concentration than allowed is detected in the fuel cell feeding. Strong interaction effects among the control loops around the last reactor, are found. Based on this, two model predictive control technologies are tested and compared in this interacted zone, in order to improve the disturbance rejection and satisfy the H2 expected quality. An exigent disturbance profile was used for simulating dynamically the complete process behavior.