A Penalty Function to Obtain Satisfactory Local Models of Complex Systems
Grzegorz Drałus, Jerzy Świątek · 2003
In this paper using a penalty function to obtain satisfactory local models in global modeling of complex systems using neural networks are discussed. Complex systems consist of several subsystems in a cascade structure. As models, non-typical neural multilayer feedforward networks were used. To formulate the global performance index the global resultant criterion, which consist a global error criterion and a penalty function for local errors, approach is proposed. Global resultant criterion required developing of a new backpropagation learning algorithm. The computer simulation results of modelling of the chosen complex system are presented.