Parameter Estimation for Batch Processes with Measurements of Large Sampling Intervals

摘要

Parameter estimation of batch processes is investigated in this paper. Due to the lack of online sensors and short length of each batch, very few off-line laboratory analysis measurements are available with a large sampling interval within each batch. To capture the nonlinear and non-Gaussian features, dual particle filters are employed to perform the state estimation and the parameter estimation in parallel. Different from most of conventional methods for the parameter estimation which only employ the measurements along the time dimension (measurements of a single batch), the measurements along the batch dimension are also taken into account in this work. In the proposed method, to avoid the estimation degeneracy due to few measurements available along the time dimension, the parameter is treated as an invariant along the time dimension and its variability is introduced along the batch dimension denoted by a random walk model. Considering that different batches share the same raw material, states among different batches are related by slowly varying or constant initial states. The application in a beer fermentation process is used to illustrate the proposed approach.