Representing the Real-Time-Behaviour of Technical Processes in Neural Nets by Using the Phase-Space-Flow of the Degrees of Freedom of the Nets

Abstract

The application of neural networks in supervision and control of technical processes does not only require the ability to classify states of process and to identify possible faulty or dangerous states but also the possibility to monitor the changes of the process-variables over time to predict eventually developing dangerous states. As the authors have shown [2] one way to store such real time behaviour in a neural net classification structure is the learning strategy of sensitisation. Meanwhile this method of teaching neural nets by means of sequences of sets of values in process which converge towards states of process, which is known to be faulty or dangerous, is successful in many applications. The mathematical background and the methods how to supervise a sensitisation were clarified last year. The first part of this paper introduces the theory of the sensitisation with a more complex mathematical background based on the phase-space-representation of the degrees of freedom of neural nets. The second part shows how a sensitisation can be supervised by the flow of trajectories of the net-parameters. Furthermore it will be shown by experimental results, how the classification behaviour of a sensitised and a non-sensitised neural net differ.