The design of parallel Kalman filters with variable forgetting factors

Author

Michael George Pappas

Date of Award

8-1986

Degree Type

Thesis

Degree Name

Master of Science

Major

Chemical Engineering

Major Professor

J. E. Doss

Committee Members

Charles Moore

Abstract

In the last decade, a number of improvements in the application of the Kalman filter to industrial state and parameter estimation problems have been made. However, there remain several unresolved performance problems which hinder the further application of these improvements.

The primary problem involves the formulation of a Kalman filter that exhibits the level of robustness required for successful application in an industrial environment. None of the existing methods provides adequate performance over the entire range of process conditions that will be experienced in industrial applications.

In this thesis, the areas where the existing methods exhibit poor performance were identified, and a new approach to Kalman filtering was developed to overcome each of these deficiencies. The result is the parallel Kalman filter, which uses two Kalman filters with the Ydstie variable forgetting factor specified with different memory lengths. The parallel filter is configured so that under any set of process conditions, one filter is specified properly. Based on the Hagglund fault detection method that was modified to use information from each component of the parallel filter, the covariance from the short memory length component filter is substituted into the long memory length component filter when a validated fault is detected. The state or parameter estimates from the long memory length filter, subject to the influence of the short memory length filter, are the estimates provided by the parallel filter algorithm. The parallel filter demonstrated the following properties:

1. Covariance windup was not experienced. The parallel filter is protected from conditions that produce covariance windup.

2. Estimate degradation due to loss of system observability was greatly reduced during exceptionally poor process conditions. Under more typical conditions, the degradation was essentially eliminated.

3. The parallel filter provides rapid and accurate response to state or parameter changes.

The Hagglund fault detection method, which uses the parameter estimate increment to determine if a true change in the parameter values is occurring, was enhanced to eliminate false alarms in the ARMA parameter estimation problem. When combined with real-time logic that observes the short memory length forgetting factor sequence, the enhanced fault detection method provides a highly reliable indicator to trigger the covariance substitution that is required to keep the parallel filter responsive to parameter changes.

Recommended Citation

Pappas, Michael George, "The design of parallel Kalman filters with variable forgetting factors. " Master's Thesis, University of Tennessee, 1986.
https://trace.tennessee.edu/utk_gradthes/13776