Students: Dissertation: Stephen Shervais

Stephen Shervais

ABSTRACT
A common problem in business is deciding on inventory and transportation policies for a physical distribution system within a changing business environment. This work addresses selection of an optimal set of policies for a multi-product, multi-echelon, multi-modal physical distribution system, in a non-stationary environment. The problem is highly multi-dimensional, even with a small system, and the fitness surface is quite often discontinuous, with low penalty and high penalty regions separated by no more than a single transport unit.

The approach used was to perform a global search for a good initial policy set using a genetic algorithm (GA) in a static environment, followed by local optimization and fitness-terrain-following in a changing environment using an adaptive critic controller design based on a Dual Heuristic Programming methodology. Performance was compared with fixed policy controllers developed using genetic algorithm and linear programming techniques.

We demonstrate a process for building a controller that will reliably improve on the performance of fixed policy controllers designed using other methodologies. Specifically, we found that the worst controller developed via this method outperformed both the LP and GA fixed policies. This process includes use of training data embodying 1/f noise, and the use of GA-derived policies as a start point for the neural controller. In addition we demonstrate the effectiveness of off-optimal, GA-developed I/O pairs as training sets for the plant model neural net, and speculate on the use of a GA as a way of testing proposed business rules.

Friday, October 6, 2000
DISSERTATION COMMITTEE
George G. Lendaris, Chair
Alan R. Raedels
Wayne W. Wakeland
Martin Zwick
Robert Fountain, Graduate Studies Rep.