Improved estimation in applications involving multiple independent goals

An estimation procedure is presented for applications where each estimated signal is subsequently used individually, e.g., for process supervision or sensor calibration, as opposed to all signals used together, e.g., for opti mization or control. The method employs multiple independent estimators in parallel, each of which is tuned to serve a single goal. It yields more acc urate estimates than the conventional procedure, does not require trial-and -error specification of relative goal weights, and is less sensitive to und er- or over-parameterization of the estimator. These advantages are demonst rated on a realistic simulation of a chemical batch reactor. (C) 2000 Elsev ier Science Ltd. All rights reserved.