Chaos control by using Motor Maps.

In this paper a new method for chaos control is proposed, consisting of an unsupervised neural network, namely a Motor Map. In particular a feedback entrainment scheme is adopted: a chaotic system with a given parameter set generates the reference trajectory for another chaotic system with different parameters to be controlled: the Motor Map is required to provide the appropriate time-varying gain value for the feedback signal. The state of the controlled system is considered as input to the Motor Map. Particular efforts have been paid to the feasibility of the implementation. Indeed, the simulations performed have been oriented to design a Motor Map suitable for an hardware realization, thus some restrictive hypotheses, such as for example a low number of neurons, have been assumed. A huge number of simulations has been carried out by considering as system to be controlled a Double Scroll Chua Attractor as well as other chaotic attractors. Several reference trajectories have also been considered: a limit cycle generated by a Chua's circuit with different parameters values, a double scroll Chua attractor, a chaotic attractor of the family of the Chua's circuit attractors. In all the simulations instead of controlling the whole state space, only two state variables have been fed back. Good results in terms of settling time (namely, the period in which the map learns the control task) and steady state errors have been obtained with a few neurons. The Motor Map based adaptive controller offers high performances, specially in the case when the reference trajectory is switched into another one. In this case, a specialization of the neurons constituting the Motor Map is observed: while a group of neurons learns the appropriate control law for a reference trajectory, another group specializes itself to control the system when the other trajectory is used as a reference. A discrete components electronic realization of the Motor Map is presented and experimental results confirming the simulation results are shown. (c) 2002 American Institute of Physics.