Abstract

The effective automatic control of Advanced Life Support Systems (ALSS) is a crucial component in space exploration. An ALSS is typically a coupled dynamical system: deterministic components that when coupled cause the overall behavior of the system very difficult to predict. We have implemented an ALSS simulator that models the basic processes occurring in the Bioregenerative Planetary Life Support System Test Complex (BIO-Plex) developed by NASA Johnson Space Center (Barta, Castillo & Fortson 1999, Kortenkamp, Bonasso & Subramanian 2001, Tri 1999). In this technical report, we present a mathematical description of our simulator and mathematical definitions of the optimization goals, the control strategies, and fitness functions used in previous work (Wu & Garibay 2003). We report findings on using a Genetic Algorithm (GA) and a Stochastic Hill-climber to learn control parameters for an ALSS simulation. We examine the behavior of these algorithms using different control strategies, fitness functions, population sizes, mutation rates, and most importantly, different representations: binary representation and three variants of the proportional representation (Wu & Garibay 2002) The GA using a proportional representation yields the best results throughout a wide range of settings for optimizing ALSS. Keywords: Genetic algorithm, optimization, life support system control, resource allocation, proportional representation, evolutionary computation. 1 1 ALSS Mathematical Description Informally, an ALSS simulation (Definition 1) is a coupled dynamical system with a set of internal simulation states (Definition 3), a control strategy (Definition 5) and a transition function (Definitions 6 to 10). The next step of a simulation (Definition 11) is calculated deterministically usi...

Citations