Development of an intelligent controller using Genetic Programming

Very complex systems require sophisticated controllers in order to achieve high performance. The demand for ever higher performance is putting the traditional controllers in crisis especially when there is a lack of information on the environment in which the system is operating or when the goals are not completely clear. These needs can be met by Intelligent Control (IC), a class of control techniques whose development is recently increased thanks to modern processing capacity. Flexibility is the key for intelligent controllers in order to adapt to different scenarios. Therefore in this master thesis an intelligent control system was designed by using Genetic Programming (GP) algorithm in Python. The capabilities of genetic programming to generate a proper controller structure are shown in off-line process, where disturbances are not taken into account and the control structure is not updated over time. In addition the behaviours of several controllers designed by genetic programming are compared in presence of disturbances in order to understand the robustness. Then an intelligent adaptive architecture has been proposed. In this instance the update of the control structure is done by using genetic programming in on-line adaptation process when a disturbance occurs in the plant. The idea of the suggested method has been tested on mass-spring-damper system and on a modified version of the standard Goddard problem. In these problems the genetic programming receives as inputs the state of the plant so that it generates the right values of control variables required in order to minimize the discrepancies over the time between the current and desired values of the state variables. The obtained results show the potentialities of genetic programming to adapt the structure of the controllers for several conditions in which plant parameters have been changed.