Data-driven ALS-SVR-ARMA2K modelling with AMPSO parameter optimisation for a high consistency refining system in papermaking.

In this study, an innovative data-driven non-linear dynamical system modelling method, ALS-SVR-ARMA2K, is proposed for a high consistency refining (HCR) process in papermaking to online estimate the freeness index. First, to completely capture the non-linear dynamics of the process, the autoregressive moving average (ARMA) model is constructed by modelling system output as a non-linear dynamical function of the past system inputs (moving average) and past system output (autoregression). Then, the non-parametric kernel learning method, least squares support vector regression (LS-SVR), is presented for learning of the ARMA model. Moreover, to increase the modelling accuracy and make the model has both good interpolation and extrapolation abilities, improvements of multi-kernel learning and adaptive output regulation are implemented. Finally, to further enhance the modelling accuracy and efficiency of the constructed ALS-SVR-ARMA2K model, the adaptive mutation particle swarm optimisation (AMPSO) is proposed to optimise the hybrid model parameters. Industrial experiments and comparative studies have been carried out on an industrial HCR process, where it has been demonstrated that the developed AMPSO-ALS-SVR-ARMA2K model produces a better modelling accuracy, stronger generalisation capability and sparsity, and faster learning speed than the other methods. Moreover, these superiorities are especially true in the case of small samples. [ABSTRACT FROM AUTHOR]