Integrated OBF-NN models for extrapolation enhancement in conventional neural networks for nonlinear systems

In this paper the integration of linear and nonlinear models in parallel for nonlinear system identification is investigated. A residuals-based sequential identification algorithm using parallel integration of linear Orthornormal basis filters (OBF) and a nonlinear feedforward (MLP) NN model is used and applied to the nonlinear Van de Vusse reactor. Results show improved extrapolation capability of the proposed method in comparison to conventional MLP NN, and opens up a promising area for further research and analysis.