Gray-box identification of dynamic models for the bleaching operation in a pulp mill

Model selection approaches for non-linear system identification: a review

“…Finally, some other linear-in-the-parameters (greybox) models have also been explored, where the model structure, non-linear model terms or network activation functions are mostly system specific (Lorito 1999;Li et al 2004;Raghavan et al 2005;Bohlin 2006;Bohlin 2006;. For example, non-linear system identification was successfully applied to the environmentally important topic of NO x emissions modelling and prediction of a 200-MW coal-fired power generation plant (Li et al 2004).…”

Identification of dual‐rate sampled systems with time delay subject to load disturbance

“…A polynomial approach was developed to represent a dual‐rate sampled system by identifying a higher‐dimensional model with the lifting technique [5], which was also involved with redundant parameters undesired for estimation. By using an output estimation in combination with interpolation between the known or observed data, a dual‐rate identification method was presented in [6], where it was pointed out arbitrary interpolation could not be allowed in particular for slow output sampling. For identifying the model parameters of an underlying single‐rate system subject to stochastic noise in the dual‐rate sampled data, an auxiliary model was used to estimate the unknown noise‐free output [7].…”

Stochastic gradient based iterative identification algorithm for a class of dual-rate Wiener systems

“…Recently, Chen et al proposed a modified SG algorithm and a multi-innovation SG algorithm for the dualrate Hammerstein system with preload nonlinearity in [24] and [25], respectively. On the other hand, multirate sampleddata systems were treated as missing data systems, and the expectation maximization (EM) algorithm was employed to estimate the parameters [26][27][28]. However, when too many data are missing, the EM algorithm results in poor parameter estimation accuracy.…”