Title :
Neural network-based control: Multiobjective design
Author_Institution :
Sch. of Commun. & Control Eng., Southern Yangtze Univ., Jiangsu, China
Abstract :
By means of linear difference inclusion (LDI) state-space representation, a general design methodology for neural network-based control systems is extended to cover various performance indexes with stability qualification. Considering the given approximation error bound between a nonlinear dynamical system and a multi-layer neural network, the proposed methodology is formulated in a set of linear matrix inequalities (LMIs). By benefiting from the common backpropagation (BP) algorithm for NN´s training and numerical solver for LMI, the design approach shows its availability.
Keywords :
linear matrix inequalities; neural nets; neurocontrollers; nonlinear control systems; state-space methods; approximation error bound; backpropagation algorithm; linear difference inclusion state-space representation; linear matrix inequalities; multiobjective design; neural network-based control; nonlinear dynamical system; performance indexes; stability qualification; Approximation error; Control systems; Design methodology; Linear matrix inequalities; Multi-layer neural network; Neural networks; Nonlinear dynamical systems; Performance analysis; Qualifications; Stability;
Conference_Titel :
Systems, Man and Cybernetics, 2002 IEEE International Conference on
Print_ISBN :
0-7803-7437-1
DOI :
10.1109/ICSMC.2002.1176370
