Nonlinear Model Predictive Control (NMPC) for SOFC

Model predictive control (MPC) is an advanced and sophisticated control tool in
comparison to classical control tools, such as P, PI or PID methods. MPC utilizes a finite number of optimized control actions along a finite prediction horizon. Although at a specific time the MPC predicts a sequence of optimal control actions, only the first control action is applied to the system. After which a repetition of the optimization process starts at the new time step.
Determining the optimal control actions depends on the predesigned performance indices that form the cost function of the system. The performance indices, for example, could be the sum of weighted second norms of the system output deviation from a set point, the input deviation and/or the rate change of input to avoid fatigue. Simply speaking, the cost function calculates the impact of control variables along a control horizon on the system
performance along a prediction horizon. The optimal control actions along the control horizon will be the values of the input that minimize the underlying cost function, where the optimization process is subjected to system constraints to protect the system from undesired input values. It is necessary to include some important operational constraints in the optimization calculation. For the case
of SOFC operation, for instance it is important to maintain the current of the fuel cell between physically feasible limits.
A fundamental model of an SOFC has been developed that forms the basis for future predictions of SOFC output power. However, the SOFC model is highly nonlinear and the power amplitude of SOFC undergoes a gain sign change at maximum power point, which makes the implementation of classical control tools difficult especially when forcing the controller output to obey amplitude saturation of rate change saturation.
In this work, a nonlinear MPC controller shows very good tracking to the input signal of power demand with keeping the output power of the SOFC within operational constraints. Furthermore, the controller succeeded in noise and error rejection, which arise from measurement noise and model imperfection, by updating the SOFC model regarding the operating temperature and pressure changes and by updating SOFC power prediction by taking into account the error between the voltage prediction and filtered measurement of SOFC voltage.
The controller could help in operation of SOFC in distributed power generation units where there is the need to run SOFC units at precise demand levels and within safety or operational constraints.
The Matlab Symbolic Math and Optimization toolboxes are used to construct and solve the cost function, while the LabVIEW is used to simulate controller performance and to investigate the on-line tuning of controller parameters.