Sushil J. Louis
Genetic Algorithm Systems Laboratory
Department of Computer Science
University of Nevada, Reno
sushil@cs.unr.edu
-
John McDonnell, Nick Gizzi
Space and Naval Warfare Systems Center
53560 Hull Street
San Diego, CA 92152
{mcdonn, gizzi}@spawar.navy.mil
This paper presents a new approach to the problem of allocating strike force assets in a dynamic targeting environment. We develop a nonlinear programming formulation that encompasses both strike and suppression responsibilities as well as multi-target and multi-threat allocations. Our approach uses a genetic algorithm augmented with a case-based memory, containing population members from past problem solving attempts, to obtain better performance over time on sequences of similar allocation problems. The case-base acts as an associative long-term memory of problem solving experience. Rather than starting with a randomly initialized population on each new allocation problem, we periodically inject a genetic algorithm's population with appropriate cases (encoded allocation strategies) from similar, previously solved problems. Using hamming distance as a simple distance (similarity) metric for choosing appropriate cases, our experimental results demonstrate the performance gains from our approach and show that our system learns to take less time to provide quality solutions to new allocation problems as it gains experience from solving other similar allocation problems.
Keywords: Asset Allocation, Genetic Algorithms, Associative Memory, Case-based Reasoning