Machine Learning for Optimal ITAE Controller Parameters for Thermal PTn Actuators

Abstract

In control theory, the ITAE criterion (integral of time-multiplied absolute value of error) is very well suited for setting the parameters of controllers, as it uses a step response and integrates the difference between the desired and actual value weighted over time. This criterion is to be minimized when setting the controller parameters. In the state of the art, parameters as example for PID controllers are found by hand and with the help of computing or Matlab toolboxes in order to minimize the ITAE or other criterions. The method presented here uses a machine learning algorithm for the automated search for the optimal controller parameters, in order to minimize the ITAE criterion. It can even be used both, in the simulation and directly on the real system. Since PTn systems have to be regulated in many cases, these are used here as example. With the application of this method, it is possible to find the parameters either using a Simulation, or directly on the real system. In the specific system, the temperature control of a thermal actuator with a small temperature chamber was applied. In particular with thermal actuators, it is often difficult or even impossible to place the sensor directly next to the heat source. This leads to PTn plant systems. The method works for this specific example and, due to its flexibility, can be extended to a huge number of applications in control theory.