Automatic design of both topology and tuning of a common parameterized controller for two families of plants using genetic programming

This paper demonstrates that a technique of evolutionary computation can be used to automatically create the design for both the topology and parameter values (tuning) for a common controller (containing various parameters representing the overall characteristics of the plant) for two families of plants. The automatically designed controller is created by means of genetic programming using a fitness measure that attempts to optimize step response and disturbance rejection while simultaneously imposing constraints on maximum sensitivity and sensor noise attenuation. The automatically designed controller outperforms the controller designed with conventional techniques. In particular, the automatically designed controller is superior to the Astrom and Hagglund controller (1995) for all plants of both families for the integral of the time-weighted absolute error (ITAE) for a step input, the ITAE for disturbance rejection, and maximum sensitivity. Averaged over all plants of both families, the ITAE for the step input for the automatically designed controller is only 58% of the value for the conventional controller; the ITAE for disturbance rejection is 91% of the value for the conventional controller; and the maximum sensitivity for the automatically designed controller is only 85% of the value for the conventional controller. The automatically designed controller is “general” in the sense that it contains free variables and therefore provides a solution to an entire category of problems (i.e., all the plants in the two families) not merely a single instance of the problem (i.e., a particular single plant)