|
Abstract:
|
Parameter estimation for continuous-time dynamic models of chemical processes must deal with unknown initial conditions, poorly parameterized models, measurement and state disturbances, irregularly sampled observations, parameter nonlinearity and partial state measurements. An approximate maximum likelihood (AMLE) formulation has been developed to address these challenges that approximates the solution of the nonlinear stochastic differential equation models using suitable basis functions (e.g., splines). Measurement noise variance is often known, but disturbance variances are typically unknown. A more stable method for estimating disturbance variance is proposed, and the AMLE technique is illustrated by estimating parameters in two dynamic models of chemical reactor processes---a nylon reactor model and a stirred tank chemical reactor simulation.
|
