ABSTRACT
A 4-input 4-output 2nd order dynamic non-linear model was simulated using a 16-24-4 Multilayer Perceptron (MLP) as in Figure 33.
2.6.4.2. Neural network training
As in the linear modelling of section 2.6.3, two data sets were generated in the neural modelling procedure — one for training and the other for testing. The weights of the neural network were adjusted using an output (1 step ahead) prediction error square sum over the training set with the BFGS algorithm [51].
