new神经网络课4-45教案2.pptVIP

F. Input Training Neural Network The aim of training IT-NN is  to minimize the following objective function: The steepest descent direction for training network is Effectiveness of IT-NN Nonlinear System Modeling with IT-NN RBF-NN Combining IT-NN with RBF Results We can see that the IT-NN with 8 nonlinear PCs will be the best choice. Fewer PCs (7, 6 and 5) cause information lost, and more PCs (10 and 9) increase the input nodes of RBF. B. Data compression Using four simulated cell-growth curves spanning the entire range of possible cell-growth profiles, having maximum normalized cell concentrations of 0.2, 0.4, 0.6, and 0.8,respectively. Each curve consist of 675 data points, divided into 45 training examples consisting of 15 data points each, producing a total training set of 180 examples(45*4). A plot of RMS error versus compression ratio used for the determination of the optimal compression ratio. Select a 3:1 compression ratio(I.e. 15 nodes in both the input and output layers being compressed to 5 nodes in the hidden layer) for the network. Following figure shows the four cell-growth curves predicted. The average error of the normalized cell concentration for the training examples is 0.0027. To demonstrate using the network to reduce noise in the cell-growth signal, we now add random noise to the four signals used for training. The four cell-growth curves with noise corresponding cell-growth curves after data compression with a 3:1 compression ratio. C. Identification 1. Forward modelling P Learning Algorithm u yp ym d d+ + - M 2. Inverse modelling Direct method C Learning Algorithm P r u ss yp + - C Learning Algorithm P r u yp Specialized method M ym There are 20 factors that directly or indirectly affect the yields of ethylene and propylene A input-compression network … Out(1) Out(2) Out(m) subgroup (1) subnet 1 subgroup (n) … hidden layer 1 hidden layer 2 hidden layer 3 subnet n … * Attaching a Dimensionality-Reduction Network to a Bac