Evolving programs and solutions using genetic programming with application to learning and adaptive control

This paper discusses two feasibility studies of Genetic Programming (GP) to the field of control theory, GP being a method inspired from nature where the goal is to create a computer program automatically from high-level statements of problems' requirements. The first feasibility study derives from stability theory and deals with evolving a program that can solve discrete-time Lyapunov equations. The second application of GP tackles the problem of producing a self-evolved Model Reference Adaptive System (MRAS). Basic structure of the programs used in the experiments are only marginally different, yet applied to seemingly quite different problems. In the first feasibility study, it was observed that GP, beside correct usage of global variables, could also purposely arrange mathematical functions and operations in an iterative manner without being explicitly programmed for the task. In the second feasibility study, a controller was evolved for a second-order process based on a pre-defined reference model.