(483e) Surrogate Function Optimization of Hybrid Power Systems Under Uncertainty

Despite the diminishing supply of non-renewable energy sources, the number of passenger vehicles purchased continues to rise with a projected 14 million cars to be sold to U.S. consumers in 2012. Hybrid electric vehicles (HEVs) combine an internal combustion engine or fuel cell with an electric motor while using a battery for energy storage. While HEVs require substantially less fuel in the long run than conventional vehicles, their high upfront cost and expensive battery replacement cost deter many buyers. Batteries are often conservatively designed to prevent damage, an effective but expensive method for handling the uncertainty which exists in battery production. More specifically, a battery which is designed with an excessive amount of capacity and baseline power will have a larger range and be far less likely to fail over time, but will be bigger, heavier and more expensive to begin with.

An empirical battery model was simulated using Matlab and Simulink, using battery capacity and baseline power load as design variables. The objective of the work was to reduce both the upfront cost and the risk of damage to the battery by optimizing capacity and baseline power. Uncertainty was added in the simulation to modify capacity, baseline power, resistance and initial state of charge. A numerical optimization method was developed specifically to handle uncertainty in the model. The method relies on repeatedly sampling and approximating the objective function surface. The method uses a trust-region approach, allowing the current sampling region to expand and contract depending on the location of the previous iteration solution. This method was found to handle uncertain optimization problems more effectively than other existing methods and robustly provided a successful solution to the HEV optimization problem.