(471b) Identification and Control of a Tubular Solid Oxide Fuel Cell ( SOFC)

In view of its use in a grid-connected system, SOFC is expected to face large variations in power demand. Under varying load, a SOFC has to satisfy three requirements. First, it has to meet the stringent requirements of the external load (such as power, voltage etc.). At the same time, it has to maintain the efficiency of the system. Also no structural or material damage can be afforded. Coupled with significant temperature gradients and fast transients, this gives rise to a very challenging control problem.

Model predictive control (MPC) is a very efficient and robust control technique for such challenging control problems. For implementation of MPC, a model needs to be identified which can be solved in real time. A data-driven model is identified in this study. Data for model identification are generated by a model which has been validated with industrial data for steps of different magnitude and directionality in different variables for wide operating conditions. So the process nonlinearity is expected to be reflected in the data and needs to be captured by the identified model. The input signal is designed maintaining linear persistence of excitation. Several linear (such as FIR, ARX etc.) and nonlinear models (such as NAARX) with increasing complexity are compared and the best model is chosen according to the AIC values of the models. Both SISO and MIMO models are identified. Several SISO models such as Voltage-power, H2 flow-power, voltage-H2 Utilization Factor are identified. For the MIMO model, voltage and H2 flow are considered as inputs. Power and utilization factors are considered as outputs. A linear model is found to be satisfactory for most SISO cases. However, a nonlinear model such as NAARX model with more cross terms is found to improve the model performance significantly for the MIMO case. All through this work, efforts have been made to come up with the simplest, yet representative model that can be used for real-time applications.

The identified model is used for designing the controller for both the SISO and MIMO systems. A PID controller and a nonlinear Model Predictive Controller are implemented. In the presence of strong interactions between the loops, the implementation of the controllers is very challenging. For the SISO control, the PID control is found to work satisfactorily. However, large overshoots in PID performance are observed for the MIMO problem. The performance of NMPC is very good for all the criteria tested. Reasons for large overshoots in the PID performance are found by comparing the control moves of the PID in comparison to that of the NMPC. Issues related to real-time implementation and tuning of the NMPC controller will also be presented in this talk.