Modification of the method of generation of control finite-state machines with continuous actions based on training examples
Igor Buzhinsky, Sergey Viktorovich Kazakov, Vladimir I. Ulyantsev, Fedor Tsarev, Anatoly Abramovich Shalyto · Journal of Computer and Systems Sciences International · 2015
Control finite-state machines can be used in the development of reliable control systems due to their clarity and because it is possible to formally verify them. The paper deals with resolving the problem of the generation of machines that control plants with complex behavior based on training examples. The input and output actions of the machines are given by real numbers. A method for the generation of machines is proposed. It is a modification of the previously proposed approaches based on the genetic and ant colony optimization algorithms. Changes include a new way of representing machines and improving the fitness function. The method makes it possible to generate machines whose behavior is more consistent with training examples than the behavior of machines generated by the known approaches.