A Case Study of a Genetically Evolved Control System

A comparison between a genetically evolved controlled system and a standard PID controlled system is carried out. In “Method and Apparatus for Automatic Synthesis Controllers” by John Koza et al., a method for the automatic design and synthesis of controllers by means of genetic programming is proposed; several genetically evolved controllers are presented and compared with controllers designed with existing methodologies. In this paper I illustrate a possible procedure to reproduce the results claimed by Koza with regard to the two lags plant controller. The system is compared to a PID proposed in a textbook. The control problem is analysed to identify specifications, requirements and characteristics of the controlled system. The lack of some constraints and data regarding the physics of the control problem are pointed out to be initial flaws of the method. The solution is searched in a not well defined search space: inferences are drawn in order to continue the analysis. A simulation of the genetically evolved controller and the PID is carried out with Matlab. Through the data presented, the two system are compared. A substantial difference in the problem specifications seems to be partially responsible for the manifested difference of performance in favour of the genetically evolved controller. Hence a new PID controller is tuned respecting the constraints assumed for the evolved controller and a new comparison is carried out. The similar results and the comparable or better performances obtained with newly tuned PID suggest to reject the hypothesis of absolute superiority of the genetic programming controller. Though, the topic is left open to discuss the achievement of Pareto optimal classes of solutions by the genetic programming methodology with more carefully chosen control constraints.