Data-driven nonlinear control of a solid oxide fuel cell system

Abstract

Solid oxide fuel cells (SOFCs) are considered to be one of the most important clean, distributed resources. However, SOFCs present a challenging control problem owing to their slow dynamics, nonlinearity and tight operating constraints. A novel data-driven nonlinear control strategy was proposed to solve the SOFC control problem by combining a virtual reference feedback tuning (VRFT) method and support vector machine. In order to fulfill the requirement for fuel utilization and control constraints, a dynamic constraints unit and an anti-windup scheme were adopted. In addition, a feedforward loop was designed to deal with the current disturbance. Detailed simulations demonstrate that the fast response of fuel flow for the current demand disturbance and zero steady error of the output voltage are both achieved. Meanwhile, fuel utilization is kept almost within the safe region.

Keywords

solid oxide fuel cell (SOFC) / data-driven method / virtual reference feedback tuning (VRFT) / support vector machine (SVM) / anti-windup

References

[1] KnyazkinV., SoderL., CanizaresC.. Control challenges of fuel cell-driven distributed generation [C]. IEEE Power Tech Conference, 2003New YorkIEEE328-333

[2] ZhangT.-j., FengGang.. Rapid load following of a SOFC power system via stable fuzzy predictive controller [J]. IEEE Trans on Fuzzy Systems, 2009, 17(2): 357-371

[3] HuoH.-b., ZhongZ.-d., ZhuX.-j., TuH.-yong.. Nonlinear dynamic modeling for a SOFC stack by using a Hammerstein model [J]. Journal of Power Sources, 2008, 175(1): 441-446

[4] WuX.-j., ZhuX.-j., CaoG.-y., TuH.-yong.. Predictive control of SOFC based on a GA-RBF neural network model [J]. Journal of Power Sources, 2008, 179(1): 232-239

[5] JuradoF.. Predictive control of solid oxide fuel cells using fuzzy Hammerstein models [J]. Journal of Power Sources, 2006, 158(1): 245-253

[6] WangX.-r., HuangB., ChenT.-wen.. Data-driven predictive control for solid oxide fuel cells [J]. Journal of Process Control, 2007, 17(2): 103-114

[7] GuardabassiG. O., SavaresiS. M.. Virtual reference direct design method: An off-line approach to data-based control system design [J]. IEEE Trans on Automatic Control, 2000, 45(5): 954-959

[8] CampiM. C., LecchiniA., SavaresiS. M.. Virtual reference feedback tuning: A direct method for the design of feedback controllers [J]. Automatica, 2002, 38(8): 1337-1346

[9] CampestriniL., EckhardD., GeversM., BazanellaA. S.. Virtual reference feedback tuning for non-minimum phase plants [J]. Automatica, 2011, 47(8): 1778-1784

[10] SalaA.. Integrating virtual reference feedback tuning into a unified closed-loop identification framework [J]. Automatica, 2007, 43(1): 178-183

[11] KanshaY., HashimotoY., ChiuM.. New results on VRFT design of PID controller [J]. Chemical Engineering Research and Design, 2008, 86(8): 925-931

[12] ScholkopfB., SmolaA. J.Learning with kernels: Support vector machines, regularization, optimization and beyond [M], 2002Cambridge, MAMIT Press251-271

[13] NiuD.-x., WangY.-l., MaX.-yong.. Optimization of support vector machine power load forecasting model based on data mining and Lyapunov exponents [J]. Journal of Central South University of Technology, 2010, 17(2): 406-412

[14] CherkasskyV., MaY.. Practical selection of SVM parameters and noise estimation for SVM regression [J]. Neural Networks, 2004, 17(1): 113-126

[15] FlakeG. W., LawrenceS.. Efficient SVM regression training with SMO [J]. Machine Learning, 2002, 46(1/2/3): 271-290