engleski

Analysis of neural network structures for modelling multiplicity of a chemical reactor

Neural network structures for modelling multiplicities in continuous chemical reactors are analysed. The following structures are investigated: 1) neural network auto regression with exogenous inputs NNARX, 2) neural network output model NNOE, 3) neural network autoregression moving averages with exogenous inputs with the error compensation of predictions NNARMAX1, 4) neural network autoregression moving averages with exogenous inputs NNARMAX2, 5) neural networks with state space innovations NNSSIF, 6) neural network with input output linearization NNIOL. The neural structures are investigated on computer simulation of patterns obtained from nonisothermal continuous reactor with an irreversible first order reaction. The neural network models are validated by statistical analysis of the residuals for two independent sets of trained and untrained patterns.

eactor multiplicity; neural network structure; NARMA models

nije evidentirano