Adaptive Interplanetary Navigation Using Genetic Algorithms

This study illustrates an automated approach for filter tuning (via model optimization) using a genetic algorithm (GA) coupled with an extended Kaiman filter. In particular, the solar radiation pressure (SRP) model of the Mars Pathfinder (MPF) spacecraft is investigated using a three month span of tracking data during the cruise phase of the mission. The results obtained in this study are compared to the best model obtained by the MPF navigation team. The GA based approach does not require gradient information about neighboring model options, hence it is capable of examining filter models of varying structure. The GA operates on a population of individuals that are selected (initially at random) from the design space. In this study, the selected design space includes 1.44E+17 possible SRP models. Each individual selected from the design space processes the tracking data set using the filter. The basis for the GA’s fitness function is a normalized sample statistic of the output residual sequence. Using the fitness values computed for each individual, the GA selects the parent population via a tournament method. For crossover, several strategies are investigated to determine the best method for quick convergence of the GA to a near optimal solution. The results show that the GA is able to determine an SRP model with a fitness value that is ~6% better than the model selected by the MPF navigation team, and produces predicted residuals that are more stable.