Constrained optimal control applied to fuel cells and vehicle systems.

Abstract

The problem of oxygen starvation coupled with air compressor saturation in fuel cells is addressed in the first half of this work. Fast load transients on a fuel cell can result in low partial oxygen pressure which can damage the fuel cell stack. The air compressor surge can disrupt the flow of air into the cathode and negatively impact fuel cell power generation. These issues are posed as constrained control problems, and solutions are proposed using reference governor and model predictive control design techniques. First a load governor is developed which determines the maximum load variation that satisfies the constraints. Two designs for the load governor based on ideas of maximal output admissible sets and receding horizon control are presented. Main emphasis is on the former approach which facilitates reduction of online computations for linear systems. Application to the nonlinear system is addressed by introduction of a step disturbance observer. We test the load governor on the Opal-RT real-time simulation platform and show that the computational time is within computation power of today's processors. To satisfy the constraints and simultaneously match an arbitrary level of current demand, a rechargeable auxiliary power source can be used along with the fuel cell. The benefits of ultracapacitors for this purpose are explained. We show that optimal use of dual power sources can be formulated in the model predictive control (MPC) framework. MPC finds the optimum balance between the use of the fuel cell and an ultracapacitor during fast load transients and enforces bounds on the ultracapacitor's state of charge and also prevents compressor surge. The second part of this work is on speed regulation for heavy duty vehicles, with multiple retarding mechanisms and actuator constraints. The service brake and compression brake of a heavy duty vehicle are coordinated in a MPC framework for velocity regulation on downhill grades. We demonstrate that MPC regulates the velocity with minimum use of service brakes while meeting the constraints. It is shown that unknown vehicle mass can degrade velocity tracking. To address this issue, parameter estimation for heavy vehicles is studied in the last part of this work. A recursive least square scheme with multiple forgetting factors is used for simultaneous online estimation of vehicle mass and time-varying road grade. The algorithm is tested successfully with experimental data.