Determination of Optimal Costly Measurement Strategies for Prediction in Linear Stochastic Systems.
MASSACHUSETTS INST OF TECH LEXINGTON LINCOLN LAB
Pagination or Media Count:
FFERENTIAL EQUATIONS, MATRICESMATHEMATICS, DATA PROCESSING, COMPUTER PROGRAMS, ANTIMISSILE DEFENSE SYSTEMS, WHITE NOISE, OPTIMIZATION, ALGORITHMSKALMAN FILTERS, STOCHASTIC DIFFERENTIAL EQUATIONS, CONTROL, STOCHASTIC PROCESSES, TIME VARYING SYSTEMS, CONTROL THEORY, DIGITAL SIMULATION, ESTIMATION THEORYThe note presents the theoretical formulation, a general purpose digital computer algorithm, and simulation results for a class of optimization problems that arise in prediction studies. The main problem is to select at each instant of time one, out of many possible, set of measurements. Each measurement strategy has an inherent cost associated with its use. The measurements are to be used so that prediction accuracy is maximized. Hence, one seeks an optimal measurement strategy, over a finite interval of time, such that a weighted combination of prediction error and measurement cost is minimized. The theory is illustrated by considering the problem of position prediction accuracy of an accelerating target in the presence of white acceleration and jerk forces, using either a position measurement or a velocity measurement. Author
- Statistics and Probability