Adaptive classification and control-rule optimisation via a learning algorithm for controlling a dynamic system

The authors present a control-rule optimizing algorithm. They describe a learning algorithm, for controlling a dynamic system, in which an incremental version of the genetic algorithm is used to learn classification of the state-space of process control while a batch version of the genetic algorithm is used to optimize a set of control actions. The dynamic system chosen was a motor-driven cart on which a pole was mounted. The learning algorithm for controlling a cart-pole balancing system has been implemented by using a real-time parallel computation architecture.<>