(453a) Generalized Predictive Control Incorporating a Battery of Observers for a PEM Fuel Cell System

A 2004 study by Pukrushpan et al. [1] involving the modeling and control of a polymer electrolyte membrane fuel cell for automotive applications demonstrated the advantages realized by optimal control strategies for the regulation of the oxygen-excess ratio under the presence of measured stack-current disturbances. Those authors proposed an architecture consisting of a static feedforward (FF) controller, an optimal LQR (linear quadratic regulator) feedback controller, and an LQG (linear quadratic Guassian) state estimator. Although this control architecture achieves adequate dynamic responses, our analysis shows that the scheme suffers from steady-state offset errors. Furthermore, the 2004 study did not investigate the robustness of the control scheme with respect to modeling errors.

In this paper we present the design and evaluation of a generalized predictive control (GPC) scheme [2-5] that does not require the introduction of a feedforward control loop. The approach follows a systematic procedure that explicitly includes the disturbance in the control-design model, and features an integrator that can effectively eliminate the presence of offset. A baseline design is considered where the oxygen-excess ratio is measured. In addition, the practical case where the oxygen-excess ratio is not measured is treated by introducing a battery of state observers.

The performance of the GPC scheme is evaluated via computer simulation studies, and is compared with the FF/LQR/LQG literature precedent [1]. It is concluded that, in contrast to the literature scheme, the GPC controller delivers offset-free regulation under the base-case conditions. Furthermore, the GPC controller coupled with the proposed battery of estimators achieves significantly better performance than the FF/LQR/LQG control architecture for the practical case where the oxygen-excess ratio must be estimated. Finally, the GPC controller realized a superior performance when there is uncertainty on the value of the throttle-area parameter, hence demonstrating better robustness properties with respect to this particular modeling error.

[1] J. T. Pukrushpan, A. G. Stefanopoulou and H. Peng, Control of fuel cell power systems: principles, modeling, analysis and feedback design, Advances in Industrial Control. Springer-Verlag London Limited (2004).

[2] O. D. Crisalle, D. E. Seborg and D. A. Mellichamp, ?Theoretical analysis of long-range predictive controllers?, Proceedings of the American Control Conference, Pittsburgh, Pennsylvania, Vol. 1, pp. 570-576, Pittsburgh, PA (1989).

[3] D. W. Clarke, C. Mohtadi and P. S. Tufts, ?Generalized predictive control ? part I. The basic algorithm?, Automatica, Vol. 23(2), pp. 137-148 (1987).

[4] D. W. Clarke, C. Mohtadi and P. S. Tufts, ?Generalized predictive control ? part II. Extensions and interpretations?, Automatica, Vol. 23(2), pp. 149-160 (1987).

[5] D. W. Clarke and C. Mohtadi, Properties of generalized predictive control, Automatica, Vol. 25(6), pp. 859-875 (1989).