Novel approaches to nonlinear model predictive control with application to high temperature fuel cells

This dissertation focuses on development of nonlinear model predictive control theory with application to residential combined heat and power fuel cell systems. Through control oriented neural network design, novel feedforward and recurrent neural network architectures are introduced that yield analytical solutions under predictive control while preserving modeling accuracy. The resulting neural predictive control strategies provide significant computational savings over existing strategies and demonstrate the superior performance of recurrent over feedforward neural networks in a predictive control framework. Nonlinear predictive control is also addressed through the development of a multiple model predictive control strategy. The strategy uses an augmented formulation to provide a framework for incorporating different steady state operating conditions and serves as a tutorial on issues related to model bank generation, state estimation, state variable updating and calculation of model weights. Additionally, a multiple model based strategy for estimating and controlling nonlinear systems in the presence of unknown disturbances is developed. The strategy augments a single baseline model with expected disturbance types, resulting in both rejection and identification of active disturbances in the system. The multiple model predictive control strategy is used to control a steam reformer, an integral part of residential fuel cell systems with right and left half plane transmission zeros. The multiple model predictive control strategy results in a smooth transition between these portions of the operating space, allowing for full utilization of the operating space. To provide setpoints to the fuel cell system, an economic optimization strategy is developed that trades off system efficiency and component lifetime as a function of operating temperatures and current density. The economic benefit of the optimization strategy is analyzed by integrating the optimization with a module level control strategy and comparing against conventional utility based strategies. Using hourly load profiles generated from climate data, the economic optimization based fuel cell control strategy shows the potential to save households in the United States up to 48% on annual utility bills.