Evolutionary design of process controllers

The genetic programming (GP) architecture offers a potentially powerful framework for `intelligent' controller design due to its inherent matching of structure to utility. The article describes the use of GP to design discrete time single loop controllers for two different processes: a linear auto-regressive exogeneous (ARX) type system and a non-linear simulated continuous stirred tank reactor (CSTR), both with constraints on the actuator outputs. It is demonstrated that the GP methodology is capable of designing recursive controllers that, for a specific class of control objectives, offer similar performance to PID controllers on the nominal systems. The advantages and disadvantages of controllers designed in this way are also discussed.