DISTRIBUTED MODEL ADAPTIVE ESTIMATION

THIS INVESTIGATION WILL DEVELOP MULTIPLE MODEL ADAPTIVE ESTIMATION (MMAE) TECHNIQUES FOR DISTRIBUTED KALMAN FILTERS (DKFS), WITH THE GOAL OF IMPROVING FILTER ROBUSTNESS WHILE MAINTAINING NEAR-OPTIMAL ACCURACY OF MULTI-SENSOR NAVIGATION SYSTEMS. THE MMAE APPROACH PROVIDES FILTER ROBUSTNESS IN SYSTEMS WITH UNCERTAIN MODEL PARAMETERS OR PARAMETER CHANGES, TYPICALLY RELATED TO SOURCES OF PROCESS NOISE OR MEASUREMENT NOISE. HOWEVER, WITH CONVENTIONAL KALMAN FILTERS, THIS APPROACH IS IMPRACTICAL WHEN MORE THAN TWO OR THREE UNCERTAIN PARAMETERS EXIST. THE DKF METHOD PERMITS A LARGE, MULTI-SENSOR FILTER TO BE PARTITIONED INTO A SET OF SMALLER LOCAL FILTERS OPERATINGIN PARALLEDL, PLUS A MASTER FILTER PERIODICALLY COMBINING THE LOCAL FILTER SOLUTIONS. THE DKF METHOD PROVIDES OPTIMAL OR NEAR-OPTIMAL ACCURACY, REDUCE PROCESSING BURDEN, AND IMPROVE FAULT TOLERANCE. THE DKF/MMAE (DMAE) TECHNIQUES DEVELOPED HERE WILL ALLOW A LARGE NUMBER OF UNCERTAIN SENSOR PARAMETERS TO BE ACCOMMODATED, VIA LOCAL ADAPTATION OF MULTIPLE LOCAL MODELS, PLUS GLOBAL ADAPTATION AT THE MASTER LEVEL. THESE NEW TECHNIQUES WILL EMPLOY DKF PARTITIONING TO REDUCE THE TOTAL NUMBER OF MULTIPLE MODELS REQUIRED, TO REDUCE THE SIZE OF EACH MODEL-SPECIFIC FILTER, AND TO IMPROVE THE OVERALL ADAPTATION PROCESS.