93-108
Compression Of Orbital Ephemerides Using Neural Networks
Thomas J. Kelly * and Jill A. Nordwall**
Abstract
Presented in this paper is a new approach to compressing orbital ephemerides which makes use of a neural network and a conformal mapping technique. Although the method utilizes common approximating polynomials, it is significantly different from the standard methods of ephemerides compression. Linear and complex.Y transformations are used to effectively remove the Keple-rian character of the motion from the cartesian representation of the trajectory approximating polynomials are then used to compactly represent the perturbed component of the motion in which case the approximation error is introduced into quantities that are themselves small corrections to the motion further compacting of the ephemerides is achieved by constructing a multi-layered neural network to store the transformation and mapping parameters as well as the polynomial coefficient~. Due to the efficiency of the network training algorithm, these parameters which collectively span many orbital revolutions are easily mapped by the network with negligible error. This enables the individual approximating polynomials to span small time intervals, thereby increasing the overall accuracy of the method, yet with no effect on the overall storage requirement of the network. This technique is illustrated for a satellite trajectory about the Earth for both high and low values of eccentricity Estimates of execution time, accuracy, and storage requirements
*Assistant Professorof Graduate Aerospace Engineering, University of Dayton, 300 College Park, KL 304, Dayton, Ohio 45469-0227 Member AIAA
**Member of the Technical Staff. Boeing Defense and Space Group, 499 Boeing Blvd., P.O. Box 2400.02, M/S JW-21, Huntsville, Alabama 35824-6402. Member AIAA.