A New Concept for Learning Control Inspired by Brain Theory

The paper explains an unconventional learning control method based on assumptions in the literature about human problem solving and information storage in neuronal networks. The on-line learning comprises two stages: The dynamic input-output behaviour of the process to be controlled is stored stepwise in a neuron-like manner into an associative memory as a predictive process model, the control strategy planned via this model by optimization of a goal oriented performance index is then trained in the same way into a second associative memory. As a general mapping the learned behaviour is in both cases in general nonlinear, and by this such a control design is especially suited for strongly nonlinear processes. Simulations demonstrate the applicability of the new control concept.