An Incremental Constructive Layer Algorithm for Controller Design

In this paper, a systematic approach for Neural Network (NN) controller design based on an incremental constructive layer algorithm is presented. The algorithm starts by considering minimal nodes in the hidden layer and choosing a pattern from those available for initial training. Further training continues with the remaining patterns resulting in a progressive increase in the number of neurons. The proposed design is carried out in three phases, namely, training, validation and pruning. A modified Goodness Factor is proposed in the paper to aid the pruning process. Simulation studies are performed on a single link robot arm. A model reference based NN controller with minimal numbers of nodes is obtained by maintaining the system error tolerance below a specified limit. Time responses obtained for the plant output and the reference signal show a satisfactory performance.

[1]  Takio Kurita,et al.  A structural learning by adding independent noises to hidden units , 1994, Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94).

[2]  James T. Kwok,et al.  Constructive algorithms for structure learning in feedforward neural networks for regression problems , 1997, IEEE Trans. Neural Networks.

[3]  K. S. Narendra,et al.  Neural networks for control theory and practice , 1996, Proc. IEEE.

[4]  Yoshio Hirose,et al.  Backpropagation algorithm which varies the number of hidden units , 1989, International 1989 Joint Conference on Neural Networks.

[5]  K. Khorasani,et al.  Structure adaptation in feed-forward neural networks , 1994, Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94).

[6]  M. Fernandez-Redondo,et al.  On the design of constructive training algorithms for multilayer feedforward , 2002, Proceedings of the 2002 International Joint Conference on Neural Networks. IJCNN'02 (Cat. No.02CH37290).

[7]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[8]  Derong Liu,et al.  A constructive algorithm for feedforward neural networks with incremental training , 2002 .

[9]  Peter J. Gawthrop,et al.  Neural networks for control systems - A survey , 1992, Autom..