Title :
B-spline recurrent neural network and its application to modelling of non-linear dynamic systems
Author :
Chan, C.W. ; Cheung, K.C. ; Hong Jin ; Zhang, H.Y.
Author_Institution :
Dept. of Mech. Eng., Hong Kong Univ., Hong Kong
Abstract :
A new recurrent neural network based on B-spline function approximation is presented. The network can be easily trained and its training converges more quickly than that for other recurrent neural networks. Moreover, an adaptive weight updating algorithm for the recurrent network is proposed. It can speed up the training process of the network greatly and its learning speed is more quickly than existing algorithms, e.g., back-propagation algorithm. Examples are presented comparing the adaptive weight updating algorithm and the constant learning rate method, and illustrating its application to modelling of nonlinear dynamic system
Keywords :
convergence; function approximation; learning (artificial intelligence); modelling; nonlinear dynamical systems; recurrent neural nets; splines (mathematics); B-spline function approximation; B-spline recurrent neural network; adaptive weight updating algorithm; constant learning rate method; learning speed; network training convergence; nonlinear dynamic system modelling; Adaptive systems; Aerodynamics; Feedback loop; Mechanical engineering; Multi-layer neural network; Neural networks; Nonlinear control systems; Nonlinear dynamical systems; Recurrent neural networks; Spline;
Conference_Titel :
American Control Conference, 1998. Proceedings of the 1998
Conference_Location :
Philadelphia, PA
Print_ISBN :
0-7803-4530-4
DOI :
10.1109/ACC.1998.694632