An experimental evaluation of Kalman filtering

Authors

  • James C. Hamilton,

    1. Department of Chemical Engineering, University of Alberta, Edmonton, Alberta T6G2G6, Canada
    Current affiliation:
    1. Imperial Oil Enterprises Ltd., Calgary, Alberta
    Search for more papers by this author
  • Dale E. Seborg,

    Corresponding author
    1. Department of Chemical Engineering, University of Alberta, Edmonton, Alberta T6G2G6, Canada
    • Department of Chemical Engineering, University of Alberta Edmonton, Alberta T6G2G6, Canada
    Search for more papers by this author
  • D. Grant Fisher

    Corresponding author
    1. Department of Chemical Engineering, University of Alberta, Edmonton, Alberta T6G2G6, Canada
    • Department of Chemical Engineering, University of Alberta Edmonton, Alberta T6G2G6, Canada
    Search for more papers by this author

Abstract

The effectiveness of the stationary form of the discrete Kalman filter for state estimation in noisy process systems was demonstrated by simulated and experimental tests on a pilot plant evaporator. The filter was incorporated into a multivariable, computer control system and resulted in good control despite process and/or measurement noise levels of 10%. The results were significantly better than those obtained when the Kalman filter was omitted or replaced by conventional exponential filters. In this application the standard Kalman filter was reasonably insensitive to incorrect estimates of initial conditions or noise statistics and to errors in model parameters. The filter estimates were sensitive to unmeasured process disturbances. However this sensitivity could be reduced by treating the noise covariance matrices R and Q as design parameters rather than noise statistics and selecting values which result in increased weighting of the process measurements relative to the calculated model states.

Ancillary