Reinforcement learning chaos control using value sensitive vector-quantization
Sabino M. Gadaleta, Gerhard Dangelmayr · 2002
A novel algorithm for the control of complex dynamical systems is introduced that extends our previously introduced approach (1999) to chaos control by combining reinforcement learning with a modified version of the growing neural-gas vector-quantization method to approximate optimal control policies. The algorithm places codebook vectors in regions of extreme reinforcement learning values and produces a codebook suitable for efficient solution of the desired control problem.