Model predictive control for proton exchange membrane fuel cell /
In this work, two constrained Model Predictive Controllers (MPC) are employed to control the air flow rate and temperature of Proton Exchange Membrane (PEM) fuel cell system. MPC formulation is based on state space model in the discrete time domain. It includes Laguerre and exponential weight functi...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Kuala Lumpur :
Kulliyyah of Engineering, International Islamic University Malaysia,
2014
|
| Subjects: | |
| Online Access: | http://studentrepo.iium.edu.my/handle/123456789/4982 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In this work, two constrained Model Predictive Controllers (MPC) are employed to control the air flow rate and temperature of Proton Exchange Membrane (PEM) fuel cell system. MPC formulation is based on state space model in the discrete time domain. It includes Laguerre and exponential weight functions. For air flow control, the objective is to track the optimum oxygen excess ratio. Oxygen starvation, surge and choke constraints are avoided by manipulating stack current, compressor mass flow rate and supply manifold pressure. Both linear and nonlinear controllers are designed. This includes: controller based on linearized first principle model (LMPC), controller based on black box linear model (BB-LMPC), and adaptive nonlinear controller (ANMPC) based on recurrent neural network (RNN) model. System states are estimated using Kalman filter for linear model and Unscented Kalman Filter (UKF) for nonlinear model. For nonlinear MPC, a successive linearization procedure is adopted. Dual UKF is used to estimate system states and RNN weights simultaneously for adaptation. Simulation shows that the inclusion of Laguerre and exponential function reduces the computation time by one order of magnitude and alleviates numerical instability for large prediction horizon. It allows the controller performance to match the performance of infinite horizon controller, Discrete Linear Quadratic Regulator (DLQR). The choice of the manipulated input to satisfy constraints is investigated. Surge and choke avoidance are successful when compressor voltage is manipulated. While, oxygen starvation is successfully prevented by filtering stack current. This constraint strategy works well without any conflictions. For nonlinear controller, the adaptation scheme improves the prediction capability of RNN model. ANMPC provides faster response with less overshoot compared to BB-LMPC. It manages to handle all the constraints better than BB-LMPC. Stochastic robustness of the controllers is tested using Monte-Carlo simulations. The analysis provides probability of instability, settling time, and tracking performance for different uncertainty levels. ANMPC remains robust till uncertainty of 60%, while LMPC and BB-LMPC can handle uncertainty up to 40%. ANMPC is the fastest, followed by BB-LMPC, then, LMPC. For tracking performance, ANMPC achieves the smallest variation in the performance compared to others. To control the thermal behavior of PEM fuel cell, a separate LMPC is employed. The coolant pump model has a delay. The pump voltage is manipulated to control stack temperature. Results show that both air flow and temperature controllers work well together. The desired temperature and oxygen excess ratio are maintained and all constraints are satisfied without conflict. |
|---|---|
| Physical Description: | xviii, 112 leaves : ill. ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 100-106) |