A mixture-of-experts framework for adaptive Kalman filtering

Abstract
This paper proposes a modular and flexible approach to adaptive Kalman filtering using the framework of a mixture-of-experts regulated by a gating network, Each expert is a Kalman filter modeled with a different realization of the unknown system parameters such as process and measurement noise, The gating network performs on-line adaptation of the weights given to individual filter estimates based on performance, This scheme compares very favorably with the classical Magill filter bank, which is based on a Bayesian technique, in terms of i) estimation accuracy, ii) quicker response to changing environments, and iii) numerical stability and computational demands, The proposed filter bank is further enhanced by periodically using a search algorithm in a feedback loop, Two search algorithms are considered, The first algorithm uses a recursive quadratic programming approach which extremizes a modified maximum likelihood function to update the parameters of the best performing filter in the bank, This particular approach to parameter adaptation allows a real-time implementation, The second algorithm uses a genetic algorithm to search for the parameter vector and is suited for post-processed data type applications. The workings and power of the overall filter bank and the suggested adaptation schemes are illustrated by a number of examples.

This publication has 33 references indexed in Scilit: