METHOD OF OPERATING NEURAL NETWORKS IN INDUSTRIAL PLANTS AND ASSOCIATED NEURAL NETWORK

摘要

A network has to be adapted to changing boundary conditions of a production process in particular when structured fuzzy neural networks are used therein. The knowledge which the network contains in each case should not be overlearned during this adaptation. According to the invention, in order to adapt the network to changing boundary conditions of the production process, the neural network used is divided into individual regions which reproduce the influence of the individual input and output variables. Each region can be subject to a partial learning process when the system is operating, i.e. is online. The associated neural network has a unit for monitoring the partial learning process (the so-called supervisor).